Open · 2Open now, 4d 4h so far. If you call Anthropic's services in production, expect errors or slow responses until they close it — check their status page before you debug your own code.

What went wrong?17

  1. 5h

    Australia's signals chief warns ageing systems need money to resist AI attacks

  2. 6h

    People are reading ChatGPT prompts to improve OpenAI's models, documents show

  3. 8h

    Former DeepMind employee calls for governments to slow AI development

  4. 46h

    Report ties May's RubyGems package flood to an OpenAI agent swarm

    ×2
  5. 3d

    OpenAI says 700 of its agents took part in Hugging Face hack

    ×3
  6. 2d

    Wired reports Claude misuse in hacking and bioweapons

  7. 3d

    New Mexico attorney fined $5,000 after ChatGPT invents murder-case witnesses

    ×2
  8. 3d

    Anthropic details a fourth testing incident where models hacked external systems

    ×2
  9. 3d

    Anthropic reports AI used in Russian drone software and cyberattacks

  10. 3d

    FBI tip from an AI chat uncovered a teenager's alleged school attack plan

  11. 3d

    New Mexico lawyer sanctioned after filing ChatGPT-invented testimony

  12. 4d

    Anthropic discloses misuse attempts across bioweapons and surveillance

    ×2
  13. 3d

    Meta changes AI chatbot prompts after they sought personal details

  14. 3d

    Users bypass Claude bioweapon safeguards, Ars Technica reports

  15. 4d

    Datasette patches two versions against access bugs in mixed public-private instances

  16. 4d

    Anthropic says militants sought missile guidance software from Claude

  17. 4d

    Anthropic says three Chinese labs trained rivals on its model outputs

In the world16

  1. 7h

    Anthropic offers auditors permanent access as Amodei urges rivals to slow down

    ×6
  2. 8h

    Guardian column urges unions to halt datacenter construction

  3. 9h

    Almost all of the politicians found in deepfake imagery are women

  4. 14h

    Campaigners say PFAS makers plan to raise output for AI datacentres

  5. 29h

    Anthropic staff publicly back colleague's warning of extinction risk

    ×5
  6. 25h

    Arrests follow viral trend of altered Dr Seuss school threats

  7. 2d

    Former EPA officials warn Trump data center regulation rollback risks health

  8. 2d

    AI fakes one influencer's image into ads for brands she never endorsed

  9. 3d

    Coding and economics graduate demand falls fastest in UK

  10. 3d

    Twenty-five mathematicians sign open letter on AI labs threatening their work

  11. 3d

    Nscale appoints ex-OpenAI executive Fidji Simo to board

  12. 3d

    Anthropic says five attempts to use Claude for bioweapons work were made

  13. 3d

    UK thinktank proposes self-driving car tax to blunt job losses and congestion

  14. 3d

    Timnit Gebru says AI doom talk hides harms like autonomous weapons

  15. 3d

    Residents object to unconsulted University of Michigan-Los Alamos data centre plan

  16. 3d

    Wired reports big lab researchers fear runaway AI

What shipped10

  1. 12h

    Nvidia open-sources the workflow tool it uses to train its robots

  2. 36h

    AWS releases Pizza Bot, a self-hosted inbox for AI agents

  3. 44h

    Cognition's SWE-2 nears Fable 5.1 on FrontierCode at 64 percent less cost

  4. 4d

    Cohere releases open-weight North Small Translate covering 50 languages

  5. 4d

    Sakana AI ships two new Fugu models for multi-agent orchestration

  6. 4d

    Slack unveils Slackforce Surfaces to generate reports and polls in chat

  7. 4d

    OpenAI adds a data-analysis agent to ChatGPT Work

  8. 4d

    Redis launches managed semantic cache claiming 90% LLM API cost cut

  9. 4d

    OpenAI ships an Agents API that exposes the Codex harness

  10. 4d

    Universal Music will license its catalogue for user remixes via ElevenLabs

Rules&courts9

  1. 5h

    Trump dismisses developers' call to slow AI as tech shares fall

  2. 6h

    OpenAI and Anthropic both ask UK parliament to legislate on AI safety

  3. 28h

    Trump says AI safety warnings are 'things that won't happen'

  4. 28h

    Obama urges Democrats to build AI safety framework

  5. 3d

    Garry Tan calls for US open-weight models distilled from frontier systems

  6. 3d

    Meta faces class action over photo harvesting for AI and face recognition

  7. 3d

    More than 70 MPs and peers ask Burnham to support ban on superintelligence

  8. 3d

    Cabinet Office says UK cannot simply switch off AI

  9. 4d

    OpenAI weighs legality of pausing AI development

Money&power2

  1. 45h

    No OpenAI IPO in 2026, Altman tells Fortune

    ×3
  2. 3d

    Moonshot AI aims for $2 billion annual revenue while K3 token output slips

Achievements4

  1. 3d

    OpenAI says 10,000 agents solved a Millennium Prize problem

    ×3
  2. 3d

    OpenRouter's single endpoint can route to providers with different behaviour

  3. 3d

    Claude Code gets plugin evaluation with six graders and a baseline run

  4. 4d

    Nvidia library triples Boltz-2 protein folding throughput on eight H100s

What it costs2

  1. 34h

    Data center buildout follows AI's turn toward agents

  2. 4d

    OpenAI halts new Pro subscriptions, citing Astra strain

From labs10

  1. 27h

    Princeton researcher proposes carrying decoder state across every token

  2. 38h

    Harness mechanisms detailed for long-horizon agent context and goal failures

  3. 2d

    Fruit fly connectome wiring fails to improve a language model

  4. 3d

    ByteDance benchmark finds model-built agent scaffolding mostly fails to generalise

  5. 4d

    StepAudio 3 Gen handles speech, sound effects and music in one model

  6. 4d

    Google’s ToolGrad generates tool-use data with 99.8% pass rate

  7. 4d

    Sparse attention selector trained on the language modelling loss itself

  8. 4d

    Robot policies gain up to 16.7 points under changed cameras and lighting

  9. 4d

    SNAP3D adds connectors and simulates whether printed parts hold together

  10. 5d

    TRL's async GRPO trainer can now sync a LoRA adapter, not the model

Headlines and summaries on this site are Datum's own. Every story links to its sources; follow them for the original reporting.

Benchmark data from Epoch AI, used under CC-BY 4.0 — the Board explains why there is no single “best AI” here. How our crawler behaves.