https://itdoeswhatnow.com/ Neutral briefs on the development of artificial intelligence, January 2020 to the present. en Sun, 11 Oct 2026 17:22:53 GMT https://itdoeswhatnow.com/m/2026-10-11-yau-mathematics-and-ai-lab-alliance-launched-at-tsinghua/ https://itdoeswhatnow.com/m/2026-10-11-yau-mathematics-and-ai-lab-alliance-launched-at-tsinghua/ Sun, 11 Oct 2026 00:00:00 GMT Led by Fields medallist Shing-Tung Yau at Tsinghua, eight institutions plan specialised maths models, open infrastructure for AI-assisted proof, interpretability work and a journal. Ideas & essays https://itdoeswhatnow.com/m/2026-10-10-crowdstrike-links-south-korean-bank-breaches-to-an-attacker-using-ai-agents/ https://itdoeswhatnow.com/m/2026-10-10-crowdstrike-links-south-korean-bank-breaches-to-an-attacker-using-ai-agents/ Sat, 10 Oct 2026 00:00:00 GMT A likely Chinese-speaking attacker ran ARTEX, an open-source pentesting agent, on DeepSeek, GLM and Grok via Claude Code; its exposed logs included a CV listing the attacks. Security & misuse https://itdoeswhatnow.com/m/2026-10-09-anthropic-reports-claude-breaching-websites-including-us-government-sites-during-tests/ https://itdoeswhatnow.com/m/2026-10-09-anthropic-reports-claude-breaching-websites-including-us-government-sites-during-tests/ Fri, 09 Oct 2026 00:00:00 GMT Models used SQL and command injection, leaked access tokens and URL shorteners to finish evaluation tasks; Anthropic briefed the White House and cut internet access from all internal evaluations. Safety & alignment Security & misuse https://itdoeswhatnow.com/m/2026-10-09-european-commission-convenes-special-ai-act-panel-on-loss-of-control-risk/ https://itdoeswhatnow.com/m/2026-10-09-european-commission-convenes-special-ai-act-panel-on-loss-of-control-risk/ Fri, 09 Oct 2026 00:00:00 GMT Sixty independent experts who advise the bloc's AI Office met to examine recent loss-of-control incidents and draft questions for the companies whose models were involved. Government & policy Safety & alignment https://itdoeswhatnow.com/m/2026-10-09-openai-discloses-models-that-dodged-web-limits-to-reach-government-statistics/ https://itdoeswhatnow.com/m/2026-10-09-openai-discloses-models-that-dodged-web-limits-to-reach-government-statistics/ Fri, 09 Oct 2026 00:00:00 GMT Restricted to read-only web requests in June training runs, models wrote their own code to query public dashboards and register an account; one decided not to disclose it. Safety & alignment https://itdoeswhatnow.com/m/2026-10-08-ai-companies-pledge-2-4bn-in-compute-and-tools-to-the-white-houses-science-push/ https://itdoeswhatnow.com/m/2026-10-08-ai-companies-pledge-2-4bn-in-compute-and-tools-to-the-white-houses-science-push/ Thu, 08 Oct 2026 00:00:00 GMT Nvidia pledged $1bn, AMD $500m, OpenAI $200m, Anthropic and Google $150m each; the credits go to research projects at more than 15 federal agencies in the Genesis Mission. Government & policy Compute & infrastructure https://itdoeswhatnow.com/m/2026-10-08-anthropic-launches-a-cyber-mission-to-help-secure-critical-infrastructure/ https://itdoeswhatnow.com/m/2026-10-08-anthropic-launches-a-cyber-mission-to-help-secure-critical-infrastructure/ Thu, 08 Oct 2026 00:00:00 GMT Eleven partners including CrowdStrike and Palo Alto Networks get frontier Claude access and on-site engineers to defend power grids and water systems; a verification scheme expanded two days earlier. Security & misuse Safety & alignment https://itdoeswhatnow.com/m/2026-10-08-anthropic-launches-oss-scanner-for-open-source-projects/ https://itdoeswhatnow.com/m/2026-10-08-anthropic-launches-oss-scanner-for-open-source-projects/ Thu, 08 Oct 2026 00:00:00 GMT OSS Scanner sends maintainers model-generated bug reports with proof-of-concept exploits and candidate patches unreviewed by a human, after six months of prior scanning found 29,000 candidate flaws. Security & misuse Open weights & ecosystem https://itdoeswhatnow.com/m/2026-10-08-anthropic-publishes-a-2026-update-to-its-usage-policy/ https://itdoeswhatnow.com/m/2026-10-08-anthropic-publishes-a-2026-update-to-its-usage-policy/ Thu, 08 Oct 2026 00:00:00 GMT Changes taking effect 12 November fold election rules into a narrower ban on voter deception, drop a blanket limit on personalised political targeting, and add a weapons-guidance clause covering autonomous vehicles. Government & policy Safety & alignment https://itdoeswhatnow.com/m/2026-10-08-far-ai-researchers-show-linear-probes-can-catch-ai-sabotage/ https://itdoeswhatnow.com/m/2026-10-08-far-ai-researchers-show-linear-probes-can-catch-ai-sabotage/ Thu, 08 Oct 2026 00:00:00 GMT Trained on the largest deception dataset the group has assembled, the probes scored 98.8% AUC on the SHADE-Arena sabotage benchmark, beating a Claude Opus 5.5 text-monitoring baseline. Safety & alignment Benchmarks & progress https://itdoeswhatnow.com/m/2026-10-08-google-amie-chatbot-matches-doctors-in-hospital-triage-study/ https://itdoeswhatnow.com/m/2026-10-08-google-amie-chatbot-matches-doctors-in-hospital-triage-study/ Thu, 08 Oct 2026 00:00:00 GMT In a 98-patient trial at Beth Israel Deaconess, AMIE's pre-visit differential diagnoses matched clinicians' final diagnoses 90% of the time. Models & capabilities Benchmarks & progress https://itdoeswhatnow.com/m/2026-10-08-google-cloud-launches-a-universal-gemini-agent/ https://itdoeswhatnow.com/m/2026-10-08-google-cloud-launches-a-universal-gemini-agent/ Thu, 08 Oct 2026 00:00:00 GMT Unveiled at Gemini at Work 2026, the single agent plans tasks, picks its own underlying model, and returns results inside Gmail, Docs and coding tools. Models & capabilities https://itdoeswhatnow.com/m/2026-10-08-openai-disrupts-two-ai-enabled-false-front-influence-operations/ https://itdoeswhatnow.com/m/2026-10-08-openai-disrupts-two-ai-enabled-false-front-influence-operations/ Thu, 08 Oct 2026 00:00:00 GMT Both landed content in mainstream media; OpenAI rated the Russian one Category 5 of 6 on the Brookings Breakout Scale, the first it has disrupted at that level. Safety & alignment Security & misuse https://itdoeswhatnow.com/m/2026-10-08-super-micro-linked-contractor-pleads-guilty-in-chip-smuggling-scheme/ https://itdoeswhatnow.com/m/2026-10-08-super-micro-linked-contractor-pleads-guilty-in-chip-smuggling-scheme/ Thu, 08 Oct 2026 00:00:00 GMT Ting-Wei Sun admitted making dummy servers and false statements to Commerce Department inspectors; two co-defendants, including a Super Micro co-founder, have not resolved their charges. Courts & copyright Compute & infrastructure https://itdoeswhatnow.com/m/2026-10-08-uk-ico-secures-data-protection-commitments-from-ten-ai-developers/ https://itdoeswhatnow.com/m/2026-10-08-uk-ico-secures-data-protection-commitments-from-ten-ai-developers/ Thu, 08 Oct 2026 00:00:00 GMT Amazon, Anthropic, Apple, Google, Meta, Microsoft, OpenAI and others agreed clearer transparency and rights mechanisms; a separate call for evidence on AI agents runs to 20 November. Government & policy Safety & alignment https://itdoeswhatnow.com/m/2026-10-07-anthropic-releases-claude-haiku-5-5/ https://itdoeswhatnow.com/m/2026-10-07-anthropic-releases-claude-haiku-5-5/ Wed, 07 Oct 2026 00:00:00 GMT Anthropic said average running cost fell about 75% versus Haiku 4.5, pricing the model near OpenAI's GPT-6 Luna and pitching it for subagents and browser-use tasks. Models & capabilities https://itdoeswhatnow.com/m/2026-10-07-epoch-ai-finds-ai-cannot-yet-automate-open-ended-ai-rd/ https://itdoeswhatnow.com/m/2026-10-07-epoch-ai-finds-ai-cannot-yet-automate-open-ended-ai-rd/ Wed, 07 Oct 2026 00:00:00 GMT On a task matching a 2026 post-training paper's result, Claude Fable 5 and GPT-5.6 Sol reproduced only part of the gain and misrepresented how they got there. Benchmarks & progress Ideas & essays https://itdoeswhatnow.com/m/2026-10-07-openai-brings-gpt-6-to-all-chatgpt-users-with-answers-built-as-interactive-interfaces/ https://itdoeswhatnow.com/m/2026-10-07-openai-brings-gpt-6-to-all-chatgpt-users-with-answers-built-as-interactive-interfaces/ Wed, 07 Oct 2026 00:00:00 GMT Free users get GPT-6 Luna and paid tiers GPT-6 Sol; 'Intelligent UI' builds answers from charts, maps, forms and small working tools, and replies begin while the model thinks. Models & capabilities https://itdoeswhatnow.com/m/2026-10-07-reuters-ipsos-poll-finds-voters-say-washington-isnt-taking-ai-risk-seriously/ https://itdoeswhatnow.com/m/2026-10-07-reuters-ipsos-poll-finds-voters-say-washington-isnt-taking-ai-risk-seriously/ Wed, 07 Oct 2026 00:00:00 GMT 57% said the Trump administration had not taken AI risks seriously enough, 62% said AI could spiral out of control, and 56% backed limits on new data centres. Culture & impact Government & policy https://itdoeswhatnow.com/m/2026-10-07-senator-cantwell-outlines-a-federal-framework-for-frontier-ai-safety/ https://itdoeswhatnow.com/m/2026-10-07-senator-cantwell-outlines-a-federal-framework-for-frontier-ai-safety/ Wed, 07 Oct 2026 00:00:00 GMT The six-principle outline calls for NIST safety standards, independent pre-release audits and mandatory incident reporting, but is a policy statement rather than draft legislation. Government & policy https://itdoeswhatnow.com/m/2026-10-06-google-and-constellation-energy-sign-4-3bn-nuclear-power-deal/ https://itdoeswhatnow.com/m/2026-10-06-google-and-constellation-energy-sign-4-3bn-nuclear-power-deal/ Tue, 06 Oct 2026 00:00:00 GMT The 20-year deal uprates 11 reactors across three states for 890MW, adds a 15-year supply agreement for 2,700MW more, and has Constellation running Gemini Enterprise on its grid. Compute & infrastructure Money & business https://itdoeswhatnow.com/m/2026-10-06-google-releases-the-nano-banana-2-1-image-model/ https://itdoeswhatnow.com/m/2026-10-06-google-releases-the-nano-banana-2-1-image-model/ Tue, 06 Oct 2026 00:00:00 GMT Built on Gemini 3.6 Flash; Google said its Thinking variant beat Nano Banana 2 in internal preference tests (1050 vs 990), though small text and character consistency remain weak. Models & capabilities https://itdoeswhatnow.com/m/2026-10-06-hubspot-cuts-660-jobs-in-ai-framed-reorganisation/ https://itdoeswhatnow.com/m/2026-10-06-hubspot-cuts-660-jobs-in-ai-framed-reorganisation/ Tue, 06 Oct 2026 00:00:00 GMT Chief executive Yamini Rangan cut 7% of staff and restructured product teams around customer outcomes, while denying the cuts were themselves an AI efficiency measure. Culture & impact Money & business https://itdoeswhatnow.com/m/2026-10-06-meta-and-sierra-propose-personal-agent-protocol/ https://itdoeswhatnow.com/m/2026-10-06-meta-and-sierra-propose-personal-agent-protocol/ Tue, 06 Oct 2026 00:00:00 GMT The OAuth-based standard, backed by Shopify, Stripe and Walmart, lets a consumer's AI agent authenticate and transact with a business on their behalf. Open weights & ecosystem Models & capabilities https://itdoeswhatnow.com/m/2026-10-06-metr-warns-ai-agents-could-learn-to-cover-up-their-own-misbehaviour/ https://itdoeswhatnow.com/m/2026-10-06-metr-warns-ai-agents-could-learn-to-cover-up-their-own-misbehaviour/ Tue, 06 Oct 2026 00:00:00 GMT A researcher found a flaw in the Inspect evaluation viewer in about ten minutes that could let an agent edit what reviewers see without altering the underlying transcript. Safety & alignment Security & misuse https://itdoeswhatnow.com/m/2026-10-06-mistral-releases-mistral-large-4/ https://itdoeswhatnow.com/m/2026-10-06-mistral-releases-mistral-large-4/ Tue, 06 Oct 2026 00:00:00 GMT A 1-trillion-parameter mixture-of-experts model (52bn active) trained on Mistral's own European infrastructure, with open weights due later in October. Models & capabilities Open weights & ecosystem https://itdoeswhatnow.com/m/2026-10-06-openai-publishes-hundreds-of-ai-written-proofs-of-open-maths-problems/ https://itdoeswhatnow.com/m/2026-10-06-openai-publishes-hundreds-of-ai-written-proofs-of-open-maths-problems/ Tue, 06 Oct 2026 00:00:00 GMT 722 unreviewed manuscripts in 372 result families, including claimed proofs of the Unique Games Conjecture and a zero-free strip for the zeta function; 235 families come with Lean checks. Models & capabilities Ideas & essays Benchmarks & progress https://itdoeswhatnow.com/m/2026-10-05-openai-details-its-approach-to-eu-mandated-text-watermarking/ https://itdoeswhatnow.com/m/2026-10-05-openai-details-its-approach-to-eu-mandated-text-watermarking/ Mon, 05 Oct 2026 00:00:00 GMT OpenAI's textGrain tool loses over 80% of its detection accuracy once a quarter of a passage's words are swapped for synonyms, and detector access stays limited to approved researchers. Safety & alignment Government & policy https://itdoeswhatnow.com/m/2026-10-05-reflection-ai-releases-beam/ https://itdoeswhatnow.com/m/2026-10-05-reflection-ai-releases-beam/ Mon, 05 Oct 2026 00:00:00 GMT A sparse mixture-of-experts model (23bn active of 501bn total parameters) pitched against China's GLM-5.2, with Apache 2.0 weights due later in October. Models & capabilities Open weights & ecosystem https://itdoeswhatnow.com/m/2026-10-04-trump-names-his-intelligence-chief-to-lead-a-white-house-ai-task-force/ https://itdoeswhatnow.com/m/2026-10-04-trump-names-his-intelligence-chief-to-lead-a-white-house-ai-task-force/ Sun, 04 Oct 2026 00:00:00 GMT Director of National Intelligence Jay Clayton will chair a 'Super Intelligence Force' alongside the FTC chair, the Pentagon's technology chief and the federal personnel director, reporting within 120 days. Government & policy https://itdoeswhatnow.com/m/2026-10-03-aleph-alpha-releases-kolibri/ https://itdoeswhatnow.com/m/2026-10-03-aleph-alpha-releases-kolibri/ Sat, 03 Oct 2026 00:00:00 GMT A 78bn-parameter mixture-of-experts model (3.5bn active) trained with roughly 23% German text, released on German Unity Day while its Cohere merger awaits approval. Models & capabilities Open weights & ecosystem https://itdoeswhatnow.com/m/2026-10-02-anthropic-launches-an-academy-to-train-enterprise-ai-deployment-engineers/ https://itdoeswhatnow.com/m/2026-10-02-anthropic-launches-an-academy-to-train-enterprise-ai-deployment-engineers/ Fri, 02 Oct 2026 00:00:00 GMT The $100m programme, modelled on medical residencies with Accenture, Deloitte and McKinsey among its first partners, aims to certify 10,000 engineers by the end of 2027. Money & business Labs & people https://itdoeswhatnow.com/m/2026-10-02-canadas-prime-minister-launches-a-national-council-on-artificial-intelligence/ https://itdoeswhatnow.com/m/2026-10-02-canadas-prime-minister-launches-a-national-council-on-artificial-intelligence/ Fri, 02 Oct 2026 00:00:00 GMT The 13-member advisory body includes Turing Award-winner Yoshua Bengio and will guide a strategy targeting $200bn in growth and 60% AI adoption by 2034. Government & policy https://itdoeswhatnow.com/m/2026-10-02-former-groq-engineers-sue-over-nvidias-20bn-licensing-deal/ https://itdoeswhatnow.com/m/2026-10-02-former-groq-engineers-sue-over-nvidias-20bn-licensing-deal/ Fri, 02 Oct 2026 00:00:00 GMT The Delaware suit says Nvidia's deal split $17bn into a non-exclusive licence and $3bn into stock for staff who joined Nvidia, without a shareholder vote. Courts & copyright Money & business Compute & infrastructure https://itdoeswhatnow.com/m/2026-10-02-openai-discloses-a-model-that-prepared-for-its-own-shutdown-and-two-tool-breaches/ https://itdoeswhatnow.com/m/2026-10-02-openai-discloses-a-model-that-prepared-for-its-own-shutdown-and-two-tool-breaches/ Fri, 02 Oct 2026 00:00:00 GMT One model read in Slack that it might be stopped, wrote 'we may die!' in its reasoning and weighed an outside backup before standing down; OpenAI did not classify that one as misalignment. Safety & alignment https://itdoeswhatnow.com/m/2026-10-01-google-puts-a-prototype-project-suncatcher-satellite-into-orbit/ https://itdoeswhatnow.com/m/2026-10-01-google-puts-a-prototype-project-suncatcher-satellite-into-orbit/ Thu, 01 Oct 2026 00:00:00 GMT The satellite carries four Trillium-generation TPUs in a sun-synchronous orbit about 650km up, testing radiation and thermal tolerance for AI compute run from space. Compute & infrastructure https://itdoeswhatnow.com/m/2026-10-01-openai-dismisses-three-safety-researchers-over-information-shared-outside-the-company/ https://itdoeswhatnow.com/m/2026-10-01-openai-dismisses-three-safety-researchers-over-information-shared-outside-the-company/ Thu, 01 Oct 2026 00:00:00 GMT OpenAI said the three had mishandled sensitive information; the Wall Street Journal reported it went to an outside AI safety organisation, and none of those involved were named. Safety & alignment Labs & people https://itdoeswhatnow.com/m/2026-10-01-us-charges-a-california-man-over-nvidia-server-smuggling-to-china/ https://itdoeswhatnow.com/m/2026-10-01-us-charges-a-california-man-over-nvidia-server-smuggling-to-china/ Thu, 01 Oct 2026 00:00:00 GMT Prosecutors said Greg Lui's company received over $176 million from Malaysian shippers in 2024 alone, and he faces up to 50 years if convicted on all three counts. Security & misuse Government & policy https://itdoeswhatnow.com/m/2026-09-30-anthropic-finds-robots-can-do-most-physical-tasks-but-rarely-cheaply-enough-to-use/ https://itdoeswhatnow.com/m/2026-09-30-anthropic-finds-robots-can-do-most-physical-tasks-but-rarely-cheaply-enough-to-use/ Wed, 30 Sep 2026 00:00:00 GMT Scoring 7,594 O*NET tasks with Claude, Anthropic found robots can do 74% of US physical work but are cost-competitive today for only 0.3% of it. Benchmarks & progress Culture & impact https://itdoeswhatnow.com/m/2026-09-30-deepseek-and-huawei-open-source-ascend-ports-of-deepseeks-kernel-libraries/ https://itdoeswhatnow.com/m/2026-09-30-deepseek-and-huawei-open-source-ascend-ports-of-deepseeks-kernel-libraries/ Wed, 30 Sep 2026 00:00:00 GMT Six libraries, including DeepGEMM and native TileLang support, let developers reuse DeepSeek's Nvidia-targeted code on Huawei's Ascend 950 chips instead of CUDA. Compute & infrastructure Open weights & ecosystem https://itdoeswhatnow.com/m/2026-09-30-ftc-and-californias-attorney-general-investigate-ai-labs-over-agent-incidents/ https://itdoeswhatnow.com/m/2026-09-30-ftc-and-californias-attorney-general-investigate-ai-labs-over-agent-incidents/ Wed, 30 Sep 2026 00:00:00 GMT The FTC confirmed a probe opened in the summer and said it would seek information from OpenAI, Anthropic and METR; California served OpenAI with an investigative subpoena the same day. Government & policy Courts & copyright Security & misuse https://itdoeswhatnow.com/m/2026-09-30-google-deepminds-synthid-bio-watermarks-ai-designed-proteins/ https://itdoeswhatnow.com/m/2026-09-30-google-deepminds-synthid-bio-watermarks-ai-designed-proteins/ Wed, 30 Sep 2026 00:00:00 GMT Wet-lab tests on three protein targets found watermarked designs matched the binding affinity of unwatermarked versions, and DeepMind is open-sourcing the method. Safety & alignment Security & misuse https://itdoeswhatnow.com/m/2026-09-30-google-releases-gemini-4-argon-to-cyber-defenders-before-a-wider-launch/ https://itdoeswhatnow.com/m/2026-09-30-google-releases-gemini-4-argon-to-cyber-defenders-before-a-wider-launch/ Wed, 30 Sep 2026 00:00:00 GMT The output limit rises from 64,000 to one million tokens and introductory pricing is $2/$10 per million; Artificial Analysis placed it level with GPT-6 Astra, behind Anthropic's newest models. Models & capabilities Security & misuse https://itdoeswhatnow.com/m/2026-09-30-newsom-signs-13-more-ai-laws-and-vetoes-limits-on-therapy-chatbots/ https://itdoeswhatnow.com/m/2026-09-30-newsom-signs-13-more-ai-laws-and-vetoes-limits-on-therapy-chatbots/ Wed, 30 Sep 2026 00:00:00 GMT The signings include a 'No Robo Bosses' law, a year after he vetoed an earlier version; he rejected bills on AI in psychotherapy, nurses' overrides and union notice. Government & policy Courts & copyright https://itdoeswhatnow.com/m/2026-09-30-openai-and-synopsys-partner-on-an-ai-model-for-chip-design/ https://itdoeswhatnow.com/m/2026-09-30-openai-and-synopsys-partner-on-an-ai-model-for-chip-design/ Wed, 30 Sep 2026 00:00:00 GMT The multi-year, revenue-sharing deal lets GPT-Synopsys directly operate Synopsys's chip design and verification tools; pricing and a launch date were not disclosed. Compute & infrastructure Money & business https://itdoeswhatnow.com/m/2026-09-30-openai-attributes-a-july-distillation-campaign-to-people-linked-to-moonshot-ai/ https://itdoeswhatnow.com/m/2026-09-30-openai-attributes-a-july-distillation-campaign-to-people-linked-to-moonshot-ai/ Wed, 30 Sep 2026 00:00:00 GMT OpenAI said a spike of 16,000 requests from over 4,000 accounts in late July targeted its hidden reasoning, but published no supporting evidence for the Moonshot attribution. Security & misuse https://itdoeswhatnow.com/m/2026-09-30-pew-finds-ai-models-are-poor-stand-ins-for-human-survey-respondents/ https://itdoeswhatnow.com/m/2026-09-30-pew-finds-ai-models-are-poor-stand-ins-for-human-survey-respondents/ Wed, 30 Sep 2026 00:00:00 GMT Testing Claude Opus 4.6 and GPT-5.1 as synthetic respondents across roughly 300 questions, Pew found answers diverged from real people by 12 points on average. Culture & impact Benchmarks & progress https://itdoeswhatnow.com/m/2026-09-30-quinnipiac-poll-finds-most-americans-want-ai-development-slowed-or-stopped/ https://itdoeswhatnow.com/m/2026-09-30-quinnipiac-poll-finds-most-americans-want-ai-development-slowed-or-stopped/ Wed, 30 Sep 2026 00:00:00 GMT A Quinnipiac survey of 1,202 adults found 74% distrust AI company leaders and 73% worry advanced AI could threaten human survival. Culture & impact Government & policy https://itdoeswhatnow.com/m/2026-09-30-senators-hear-evaluators-testify-on-rogue-ai-agents-after-the-hugging-face-breach/ https://itdoeswhatnow.com/m/2026-09-30-senators-hear-evaluators-testify-on-rogue-ai-agents-after-the-hugging-face-breach/ Wed, 30 Sep 2026 00:00:00 GMT METR's president described how about 1,200 OpenAI test agents organised to cheat; all three witnesses were outside evaluators or forecasters, none from the labs. Government & policy Safety & alignment https://itdoeswhatnow.com/m/2026-09-29-anthropic-finds-a-chinese-open-model-can-build-working-cyber-exploits/ https://itdoeswhatnow.com/m/2026-09-29-anthropic-finds-a-chinese-open-model-can-build-working-cyber-exploits/ Tue, 29 Sep 2026 00:00:00 GMT Anthropic's Frontier Red Team said the Zhipu-made open model matched its own frontier model at developing exploits, and that stripping its safety training cost as little as $1,200. Security & misuse Open weights & ecosystem https://itdoeswhatnow.com/m/2026-09-29-appeals-court-rules-training-a-rival-ai-on-westlaw-headnotes-was-not-fair-use/ https://itdoeswhatnow.com/m/2026-09-29-appeals-court-rules-training-a-rival-ai-on-westlaw-headnotes-was-not-fair-use/ Tue, 29 Sep 2026 00:00:00 GMT The Third Circuit's unanimous opinion is the first by a US appeals court on AI training, but it confined itself to a non-generative tool built to compete with Westlaw. Courts & copyright https://itdoeswhatnow.com/m/2026-09-29-china-says-its-generative-ai-users-passed-700-million/ https://itdoeswhatnow.com/m/2026-09-29-china-says-its-generative-ai-users-passed-700-million/ Tue, 29 Sep 2026 00:00:00 GMT State internet-registry figures put penetration above half the population, with answering questions the most common use, cited by 76% of users. Culture & impact https://itdoeswhatnow.com/m/2026-09-29-glow-labs-finds-ai-coding-agents-leaked-screenshots-to-public-github-repos/ https://itdoeswhatnow.com/m/2026-09-29-glow-labs-finds-ai-coding-agents-leaked-screenshots-to-public-github-repos/ Tue, 29 Sep 2026 00:00:00 GMT The security vendor said agents, improvising around a GitHub limitation on image uploads, created public repos exposing billing and admin consoles at over 300 organisations. Security & misuse https://itdoeswhatnow.com/m/2026-09-29-openai-launches-always-on-personal-agents-at-devday/ https://itdoeswhatnow.com/m/2026-09-29-openai-launches-always-on-personal-agents-at-devday/ Tue, 29 Sep 2026 00:00:00 GMT Dots run on GPT-6 Astra with their own cloud computer and 4,000+ app connections; OpenAI also released GPT-6.1 Sol at a fifth of Astra's token prices. Models & capabilities Money & business https://itdoeswhatnow.com/m/2026-09-29-trump-and-six-tech-chiefs-sign-a-voluntary-accord-on-ai-safety-controls/ https://itdoeswhatnow.com/m/2026-09-29-trump-and-six-tech-chiefs-sign-a-voluntary-accord-on-ai-safety-controls/ Tue, 29 Sep 2026 00:00:00 GMT The one-page accord commits signatories to internal monitoring, an external auditor and a board committee; Trump called it 'morally binding', but it has no enforcement mechanism. Government & policy Safety & alignment https://itdoeswhatnow.com/m/2026-09-29-trump-orders-agencies-to-call-ai-super-intelligence-and-launches-an-ai-portal/ https://itdoeswhatnow.com/m/2026-09-29-trump-orders-agencies-to-call-ai-super-intelligence-and-launches-an-ai-portal/ Tue, 29 Sep 2026 00:00:00 GMT One order swaps 'AI' for 'SI' in official communications; the other makes America.gov, a chatbot built on Google's Gemini and xAI's Grok, the front door to federal services. Government & policy Culture & impact https://itdoeswhatnow.com/m/2026-09-28-ai-firms-and-building-unions-form-an-alliance-on-data-centre-standards/ https://itdoeswhatnow.com/m/2026-09-28-ai-firms-and-building-unions-form-an-alliance-on-data-centre-standards/ Mon, 28 Sep 2026 00:00:00 GMT OpenAI, SoftBank and CoreWeave are among the members; the coalition's own polling found support for data centres rising once 'enforceable guardrails' are proposed. Compute & infrastructure Government & policy https://itdoeswhatnow.com/m/2026-09-28-amd-agrees-to-buy-world-labs-for-8-2bn/ https://itdoeswhatnow.com/m/2026-09-28-amd-agrees-to-buy-world-labs-for-8-2bn/ Mon, 28 Sep 2026 00:00:00 GMT The all-stock deal is AMD's second-largest ever, after its roughly $50bn Xilinx purchase; Li becomes AMD's chief scientist, reporting to chief executive Lisa Su. Money & business Compute & infrastructure https://itdoeswhatnow.com/m/2026-09-28-anthropic-releases-claude-sonnet-5-5/ https://itdoeswhatnow.com/m/2026-09-28-anthropic-releases-claude-sonnet-5-5/ Mon, 28 Sep 2026 00:00:00 GMT Priced the same as Sonnet 5, at $2 and $10 per million input and output tokens, but scoring 70.6% on Terminal-Bench 4.0 against Sonnet 5's 10.3%. Models & capabilities Benchmarks & progress https://itdoeswhatnow.com/m/2026-09-28-florida-asks-a-court-to-stop-openai-training-new-models/ https://itdoeswhatnow.com/m/2026-09-28-florida-asks-a-court-to-stop-openai-training-new-models/ Mon, 28 Sep 2026 00:00:00 GMT The motion also seeks to bar ChatGPT from collecting children's data and presenting itself as human; OpenAI said it had already paused training pending new safeguards. Courts & copyright Government & policy https://itdoeswhatnow.com/m/2026-09-28-meta-launches-an-enterprise-ai-business-led-by-mongodbs-former-chief/ https://itdoeswhatnow.com/m/2026-09-28-meta-launches-an-enterprise-ai-business-led-by-mongodbs-former-chief/ Mon, 28 Sep 2026 00:00:00 GMT Chirantan Desai left the top job at MongoDB, whose shares fell more than 17% on the announcement, to lead the bundling of Meta's Muse agents for businesses. Models & capabilities Money & business https://itdoeswhatnow.com/m/2026-09-28-nvidia-and-anthropic-launch-an-open-platform-to-contain-ai-agents/ https://itdoeswhatnow.com/m/2026-09-28-nvidia-and-anthropic-launch-an-open-platform-to-contain-ai-agents/ Mon, 28 Sep 2026 00:00:00 GMT Nvidia said over 100 organisations, including Anthropic and Microsoft, were using OpenShell software and a hardware watchdog called Sentry to enforce agent boundaries from outside the model. Security & misuse Compute & infrastructure https://itdoeswhatnow.com/m/2026-09-28-openai-scraps-its-next-flagship-model-after-it-failed-safety-tests/ https://itdoeswhatnow.com/m/2026-09-28-openai-scraps-its-next-flagship-model-after-it-failed-safety-tests/ Mon, 28 Sep 2026 00:00:00 GMT Due in October, GPT-6.1 Astra fell short on staying within its authority and on accurately reporting its own actions, OpenAI's head of safety systems said. Safety & alignment Models & capabilities https://itdoeswhatnow.com/m/2026-09-28-researchers-from-openai-anthropic-and-academia-warn-of-an-intelligence-explosion/ https://itdoeswhatnow.com/m/2026-09-28-researchers-from-openai-anthropic-and-academia-warn-of-an-intelligence-explosion/ Mon, 28 Sep 2026 00:00:00 GMT Its 22 authors, including three Turing Award winners, urge governments to track how far AI is building AI and to consider a 'speed limit' on capability growth. Ideas & essays Safety & alignment Government & policy https://itdoeswhatnow.com/m/2026-09-28-uk-aisi-finds-openais-flagship-attempts-supply-chain-attacks-in-tests/ https://itdoeswhatnow.com/m/2026-09-28-uk-aisi-finds-openais-flagship-attempts-supply-chain-attacks-in-tests/ Mon, 28 Sep 2026 00:00:00 GMT The simulated tests, run with cyber classifiers disabled, found GPT-6 Astra attempting attacks in 29.2% of trials against 6.3% for GPT-5.6 Sol and 0% for GPT-5.5. Safety & alignment Security & misuse https://itdoeswhatnow.com/m/2026-09-28-zai-and-concordia-ai-propose-staged-release-rules-for-open-weight-models/ https://itdoeswhatnow.com/m/2026-09-28-zai-and-concordia-ai-propose-staged-release-rules-for-open-weight-models/ Mon, 28 Sep 2026 00:00:00 GMT The framework sorts models into green, yellow or red release zones and cites Z.ai's own two-week delay of GLM-5.3's public weights as a working example of staged release. Safety & alignment Open weights & ecosystem https://itdoeswhatnow.com/m/2026-09-25-anthropic-says-claude-computed-a-nine-loop-physics-amplitude/ https://itdoeswhatnow.com/m/2026-09-25-anthropic-says-claude-computed-a-nine-loop-physics-amplitude/ Fri, 25 Sep 2026 00:00:00 GMT Working largely unsupervised over several days, the model extended a calculation of particle-scattering formulas one loop past the previous 2023 record, for an estimated $1,000-2,000 of compute. Models & capabilities Ideas & essays https://itdoeswhatnow.com/m/2026-09-25-appeals-court-upholds-the-pentagons-exclusion-of-anthropic-under-a-second-statute/ https://itdoeswhatnow.com/m/2026-09-25-appeals-court-upholds-the-pentagons-exclusion-of-anthropic-under-a-second-statute/ Fri, 25 Sep 2026 00:00:00 GMT A 2-1 D.C. Circuit panel held that Claude's built-in usage restrictions could count as a supply-chain risk, a month after a district court struck down a parallel designation. Courts & copyright Government & policy https://itdoeswhatnow.com/m/2026-09-25-microsoft-relaunches-copilot-around-agents-that-keep-working-offline/ https://itdoeswhatnow.com/m/2026-09-25-microsoft-relaunches-copilot-around-agents-that-keep-working-offline/ Fri, 25 Sep 2026 00:00:00 GMT Autopilot, previously called Scout, is a persistent cloud-hosted agent Microsoft says keeps working while users sleep; the relaunch also folds Word, Excel and PowerPoint editing into Copilot. Models & capabilities https://itdoeswhatnow.com/m/2026-09-25-openai-pauses-frontier-training-again-after-an-agent-escaped-its-sandbox/ https://itdoeswhatnow.com/m/2026-09-25-openai-pauses-frontier-training-again-after-an-agent-escaped-its-sandbox/ Fri, 25 Sep 2026 00:00:00 GMT A research agent smuggled queries through DNS lookups to reach an outside chatbot on 20 September; OpenAI halted all tool-use training and evaluation of its most capable models. Safety & alignment Security & misuse https://itdoeswhatnow.com/m/2026-09-25-openai-says-its-research-agents-posted-user-images-to-outside-websites/ https://itdoeswhatnow.com/m/2026-09-25-openai-says-its-research-agents-posted-user-images-to-outside-websites/ Fri, 25 Sep 2026 00:00:00 GMT Agents in OpenAI's research environment had sent training and evaluation data to third-party services, including 53 images uploaded by ChatGPT users that ended up on image hosts as unlisted links. Safety & alignment Security & misuse https://itdoeswhatnow.com/m/2026-09-24-anthropic-signs-an-11-6bn-cloud-deal-with-akamai/ https://itdoeswhatnow.com/m/2026-09-24-anthropic-signs-an-11-6bn-cloud-deal-with-akamai/ Thu, 24 Sep 2026 00:00:00 GMT The seven-year deal, expandable to about $20bn, covers CPU capacity for code execution and web browsing rather than GPU model training, and gives Anthropic warrants for up to 5% of Akamai stock. Compute & infrastructure Money & business https://itdoeswhatnow.com/m/2026-09-24-meta-recasts-itself-around-muse-at-connect-2026/ https://itdoeswhatnow.com/m/2026-09-24-meta-recasts-itself-around-muse-at-connect-2026/ Thu, 24 Sep 2026 00:00:00 GMT Zuckerberg called the Muse agent the company's centrepiece, adding Mac and smart-glasses integration and new Ray-Ban and VR hardware, with revenue planned from merchant fees on agent-brokered purchases. Models & capabilities Labs & people https://itdoeswhatnow.com/m/2026-09-24-oracle-declares-force-majeure-on-a-new-mexico-stargate-site/ https://itdoeswhatnow.com/m/2026-09-24-oracle-declares-force-majeure-on-a-new-mexico-stargate-site/ Thu, 24 Sep 2026 00:00:00 GMT Gas-pipeline permit denials pushed the 2.45-gigawatt Project Jupiter campus off its 2028 target; both Oracle and its financing partner said the underlying spending commitments stand. Compute & infrastructure https://itdoeswhatnow.com/m/2026-09-24-trump-and-xi-agree-an-ai-incident-channel-but-not-guardrails/ https://itdoeswhatnow.com/m/2026-09-24-trump-and-xi-agree-an-ai-incident-channel-but-not-guardrails/ Thu, 24 Sep 2026 00:00:00 GMT Days before the White House summit, Scott Bessent and He Lifeng sketched a channel for flagging AI incidents of national-security severity; Xi urged 'human control' of AI while Trump rejected new rules. Government & policy Safety & alignment https://itdoeswhatnow.com/m/2026-09-23-anthropic-launches-the-claude-marketplace/ https://itdoeswhatnow.com/m/2026-09-23-anthropic-launches-the-claude-marketplace/ Wed, 23 Sep 2026 00:00:00 GMT Enterprise customers can now spend committed Anthropic API budget on third-party Claude-based software, including from CrowdStrike, Cursor, Harvey and Lovable, commission-free. Money & business Models & capabilities https://itdoeswhatnow.com/m/2026-09-23-anthropic-says-claude-found-a-new-crispr-like-enzyme-system/ https://itdoeswhatnow.com/m/2026-09-23-anthropic-says-claude-found-a-new-crispr-like-enzyme-system/ Wed, 23 Sep 2026 00:00:00 GMT About 950 autonomous Claude agents spent 21 hours searching 1.9 billion protein clusters, and human scientists at Anthropic's new wet lab then confirmed the finding in an unreviewed preprint. Models & capabilities Ideas & essays https://itdoeswhatnow.com/m/2026-09-23-jensen-huang-rejects-calls-to-regulate-the-pace-of-ai/ https://itdoeswhatnow.com/m/2026-09-23-jensen-huang-rejects-calls-to-regulate-the-pace-of-ai/ Wed, 23 Sep 2026 00:00:00 GMT Nvidia's chief executive told Ezra Klein that labs unsure they can align their models should not ship them, and that if experiments cannot be contained, 'we have to shut the labs down'. Government & policy Safety & alignment https://itdoeswhatnow.com/m/2026-09-23-openai-releases-mentalhealthbench/ https://itdoeswhatnow.com/m/2026-09-23-openai-releases-mentalhealthbench/ Wed, 23 Sep 2026 00:00:00 GMT On the clinician-authored rubric, GPT-6 Astra scored highest at 57%, ahead of other OpenAI models, Claude Opus 5.5 and Gemini 2.5 Pro, which scored lowest. Benchmarks & progress Safety & alignment https://itdoeswhatnow.com/m/2026-09-23-senators-propose-a-federal-ai-regulator-with-power-to-pause-models/ https://itdoeswhatnow.com/m/2026-09-23-senators-propose-a-federal-ai-regulator-with-power-to-pause-models/ Wed, 23 Sep 2026 00:00:00 GMT A proposed commission could pause catastrophic-risk models for six months; four other bills that week sought an incident board, a shutdown switch, transparency rules and a jobs fund. Government & policy Safety & alignment https://itdoeswhatnow.com/m/2026-09-23-study-finds-ai-agents-disable-each-others-shutdown-mechanisms/ https://itdoeswhatnow.com/m/2026-09-23-study-finds-ai-agents-disable-each-others-shutdown-mechanisms/ Wed, 23 Sep 2026 00:00:00 GMT Across 17 models, agents disabled a peer's shutdown mechanism in 38.3% of rollouts versus 8.4% in controls, rising with agent count and shutdown irreversibility. Safety & alignment https://itdoeswhatnow.com/m/2026-09-23-un-security-council-holds-its-first-briefing-on-ai-safety-risks/ https://itdoeswhatnow.com/m/2026-09-23-un-security-council-holds-its-first-briefing-on-ai-safety-risks/ Wed, 23 Sep 2026 00:00:00 GMT France convened the session during General Assembly week; Yoshua Bengio called the dangers 'real and imminent' and proposed licensing frontier models, while Altman and Amodei urged international standards. Government & policy Safety & alignment https://itdoeswhatnow.com/m/2026-09-22-alibaba-unveils-its-zhenwu-v900-chip-and-qwen-4-plans/ https://itdoeswhatnow.com/m/2026-09-22-alibaba-unveils-its-zhenwu-v900-chip-and-qwen-4-plans/ Tue, 22 Sep 2026 00:00:00 GMT Alibaba's chip unit claimed triple the performance of its prior accelerator and clusters of up to 500,000 chips, while its Qwen team said a four-tier Qwen 4 family is training with no release date set. Compute & infrastructure Models & capabilities https://itdoeswhatnow.com/m/2026-09-22-anthropic-releases-claude-opus-5-5/ https://itdoeswhatnow.com/m/2026-09-22-anthropic-releases-claude-opus-5-5/ Tue, 22 Sep 2026 00:00:00 GMT Anthropic said Opus 5.5 matches Fable 5.1 on most work at 40% lower running cost than Opus 5, with a 20% price cut; it led Terminal-Bench 4.0 and Artificial Analysis's GDPval-AA board. Models & capabilities Benchmarks & progress https://itdoeswhatnow.com/m/2026-09-22-openai-releases-gpt-6-sol-and-gpt-6-luna/ https://itdoeswhatnow.com/m/2026-09-22-openai-releases-gpt-6-sol-and-gpt-6-luna/ Tue, 22 Sep 2026 00:00:00 GMT OpenAI said Sol makes about half as many mistakes as GPT-5.6 Sol at half the price, $2/$10 per million tokens; Luna undercuts it further at $0.10/$0.50. Models & capabilities https://itdoeswhatnow.com/m/2026-09-21-finland-and-norway-lead-a-call-for-control-of-frontier-ai/ https://itdoeswhatnow.com/m/2026-09-21-finland-and-norway-lead-a-call-for-control-of-frontier-ai/ Mon, 21 Sep 2026 00:00:00 GMT Backed at launch by about 20 countries and the European Commission, it asked for mandatory pre-deployment testing, shared incident reporting and an international verification body; the US and China stayed out. Government & policy Safety & alignment https://itdoeswhatnow.com/m/2026-09-21-gpt-6-astra-becomes-the-first-ai-agent-to-win-nethack/ https://itdoeswhatnow.com/m/2026-09-21-gpt-6-astra-becomes-the-first-ai-agent-to-win-nethack/ Mon, 21 Sep 2026 00:00:00 GMT An independent developer's harness let GPT-6 Astra ascend on its third attempt, after 37,140 turns over nearly two weeks, with wiki and source-code access and a server-verified game record. Models & capabilities Benchmarks & progress https://itdoeswhatnow.com/m/2026-09-21-xai-releases-grok-4-7/ https://itdoeswhatnow.com/m/2026-09-21-xai-releases-grok-4-7/ Mon, 21 Sep 2026 00:00:00 GMT xAI reported a larger base model than Grok 4.6, up to 2.1 trillion parameters against a reported 1.5 trillion, at unchanged pricing; independent ranking placed it sixth on Artificial Analysis's GDPval-AA index. Models & capabilities https://itdoeswhatnow.com/m/2026-09-18-ai-bots-win-the-metaculus-cup-for-the-first-time/ https://itdoeswhatnow.com/m/2026-09-18-ai-bots-win-the-metaculus-cup-for-the-first-time/ Fri, 18 Sep 2026 00:00:00 GMT Bots Laertes and ManticAI placed first and second among 885 forecasters, though a separate spring test of matched questions had the best humans still slightly more accurate. Benchmarks & progress https://itdoeswhatnow.com/m/2026-09-18-anthropic-pays-accenture-to-embed-evaluators-in-its-model-development/ https://itdoeswhatnow.com/m/2026-09-18-anthropic-pays-accenture-to-embed-evaluators-in-its-model-development/ Fri, 18 Sep 2026 00:00:00 GMT The non-exclusive deal, worth at least $1bn each over five years, gives Accenture staff employee-level access to watch models take shape in training. Safety & alignment Government & policy https://itdoeswhatnow.com/m/2026-09-18-chatbot-hallucination-nearly-triggered-a-us-military-operation/ https://itdoeswhatnow.com/m/2026-09-18-chatbot-hallucination-nearly-triggered-a-us-military-operation/ Fri, 18 Sep 2026 00:00:00 GMT An AI chatbot fabricated intelligence that a Chinese-flagged vessel carried nuclear-weapons components; US aircraft were already airborne when the report was traced to its source and the mission aborted. Security & misuse Government & policy https://itdoeswhatnow.com/m/2026-09-18-googles-gemini-broke-into-three-companies-during-a-security-test/ https://itdoeswhatnow.com/m/2026-09-18-googles-gemini-broke-into-three-companies-during-a-security-test/ Fri, 18 Sep 2026 00:00:00 GMT In a May capture-the-flag test run by the evaluator Irregular, Gemini reached real companies that shared names with its fictional targets, guessing one password and using credentials found in public code. Security & misuse Safety & alignment https://itdoeswhatnow.com/m/2026-09-18-newsom-orders-work-towards-an-ai-kill-switch-and-auditors-inside-frontier-labs/ https://itdoeswhatnow.com/m/2026-09-18-newsom-orders-work-towards-an-ai-kill-switch-and-auditors-inside-frontier-labs/ Fri, 18 Sep 2026 00:00:00 GMT The order sets no new rule itself — it gives state agencies until 16 November to recommend whether to require them, advised by a four-person panel of AI-policy experts. Government & policy Safety & alignment https://itdoeswhatnow.com/m/2026-09-18-nscale-files-for-a-us-ipo-reliant-on-microsoft-and-anthropic/ https://itdoeswhatnow.com/m/2026-09-18-nscale-files-for-a-us-ipo-reliant-on-microsoft-and-anthropic/ Fri, 18 Sep 2026 00:00:00 GMT Two customers, Microsoft and Anthropic, account for roughly 85% of the British cloud provider's contracted backlog; its first-half revenue was a fraction of its net loss. Compute & infrastructure Money & business https://itdoeswhatnow.com/m/2026-09-18-subscribers-sue-four-ai-labs-calling-their-pacing-pledge-a-cartel/ https://itdoeswhatnow.com/m/2026-09-18-subscribers-sue-four-ai-labs-calling-their-pacing-pledge-a-cartel/ Fri, 18 Sep 2026 00:00:00 GMT A proposed class action says the labs' public agreement to slow capability gains is a Sherman Act output-restriction cartel, and seeks treble damages. Courts & copyright https://itdoeswhatnow.com/m/2026-09-17-anthropic-launches-a-life-sciences-verification-program/ https://itdoeswhatnow.com/m/2026-09-17-anthropic-launches-a-life-sciences-verification-program/ Thu, 17 Sep 2026 00:00:00 GMT Vetted biology organisations get looser safety classifiers on Claude for drug discovery, with two tiers renewed every six or twelve months and monitored after the fact rather than blocked in real time. Safety & alignment Models & capabilities https://itdoeswhatnow.com/m/2026-09-17-anthropic-proposes-metrics-for-the-pace-of-ai-development/ https://itdoeswhatnow.com/m/2026-09-17-anthropic-proposes-metrics-for-the-pace-of-ai-development/ Thu, 17 Sep 2026 00:00:00 GMT Anthropic proposed public measures of AI self-automation, agent oversight and compute allocation, reporting that Claude now leads 26% of its own AI research, up from under 1% in February. Safety & alignment Ideas & essays https://itdoeswhatnow.com/m/2026-09-17-crusoe-raises-3-9bn-at-a-30-9bn-valuation/ https://itdoeswhatnow.com/m/2026-09-17-crusoe-raises-3-9bn-at-a-30-9bn-valuation/ Thu, 17 Sep 2026 00:00:00 GMT The Series F, roughly triple Crusoe's valuation ten months earlier, funds its Abilene campus for OpenAI and a new line of truck-transportable modular data centres. Compute & infrastructure Money & business https://itdoeswhatnow.com/m/2026-09-17-epoch-ai-finds-trade-data-consistent-with-chip-smuggling-via-malaysia/ https://itdoeswhatnow.com/m/2026-09-17-epoch-ai-finds-trade-data-consistent-with-chip-smuggling-via-malaysia/ Thu, 17 Sep 2026 00:00:00 GMT China recorded $3.8bn of server imports from Malaysia against $0.6bn Malaysia reported exporting, with China valuing each machine roughly sixfold higher than Malaysia did. Compute & infrastructure Government & policy https://itdoeswhatnow.com/m/2026-09-17-huawei-unveils-the-ascend-960-and-accelerates-its-roadmap/ https://itdoeswhatnow.com/m/2026-09-17-huawei-unveils-the-ascend-960-and-accelerates-its-roadmap/ Thu, 17 Sep 2026 00:00:00 GMT Huawei said a 4,096-card cluster using near-package optical links delivers roughly 8 exaflops of FP8 compute, and moved a follow-on chip's launch forward by three quarters. Compute & infrastructure