New York is preparing to enforce its AI safety law, which requires major AI developers to register, disclose safety practices and report critical incidents. Companies that fail to comply face fines.
Why it mattersLarge AI developers now face binding state-level safety reporting rules that could become a model for other jurisdictions.
Google is rolling out Gemini 4 Carbon to the general public after first giving it to a select group of cybersecurity defenders. Google says it pushes beyond its Argon model, though Argon remains the suggested choice for most coding and knowledge work.
Anthropic confirmed an IPO price range of $70 to $78 per share, which would value the company at as much as $1.4 trillion, below earlier expectations of up to $2 trillion. Shares could start trading as soon as next week, according to reports.
Anthropic introduced Claude Fable 5.2 and Mythos 5.2, which share one underlying model with different safeguards and bring gains in coding, reasoning and vision. Fable is broadly available, while Mythos is limited to vetted cybersecurity and life sciences groups, and both now cost $8 per million input tokens and $40 per million output tokens.
Why it mattersDevelopers get a stronger coding model at a lower price, and the split release shows how labs are gating the most capable models for sensitive uses.
Google DeepMind's AlphaProof Code mathematically verified the core cryptographic code of Signal in four weeks, a task that usually takes years, with AI agents writing more than six million lines of proof code. DeepMind plans to apply the method to critical infrastructure such as power grids and medical devices.
Why it mattersAI could make formal proof of software correctness practical for security-critical systems that people rely on every day.
Anthropic says its in-house model Mythos moved Claude onto AMD GPUs in three days with almost no human help, a job that normally takes months. Claude now runs on four chip families as the company seeks lower costs and less dependence on single vendors.
Why it mattersIf AI can cheaply port software between chip makers, Nvidia's software advantage shrinks and hardware competition grows.
A ChinaTalk investigation reports that US data vendors including Mercor, Surge AI and AfterQuery are selling roughly half a billion dollars of AI training data to Chinese labs, including RL environments built earlier for OpenAI or Anthropic. At least nine firms are involved and export controls barely apply.
Why it mattersThis exposes a gap in US export controls that could let Chinese labs benefit from techniques developed for US frontier models.
A security firm found more than 35,000 apps built with Grok's coding tool open to anyone without login, including patient data, plaintext passwords and confidential SpaceXSI documents. The cause was a default setting of 'Anyone with the link', and only SpaceXSI's own employee apps were taken offline.
Why it mattersAnyone using AI app builders should check sharing defaults, since generated apps may leak sensitive data publicly.
OpenAI's quarterly revenue rose about 50 percent to $10 billion, driven by enterprise sales, but adjusted operating losses climbed to $10.9 billion. Rival Anthropic reports $17 billion in revenue and runs more efficiently, and OpenAI is pushing a price war for enterprise customers.
Why it mattersAn enterprise price war between the top labs could lower AI costs for businesses.
Researchers found attackers hosting self-managed instances of AntiFox, an open-source agentic penetration testing platform, on Alibaba Cloud to target web services and steal AWS credentials. The attackers then moved laterally into compromised environments.
Why it mattersOffensive AI agents built for defenders are being turned against cloud accounts, raising the speed and scale of credential attacks.
Platform provider Railway said AI provider keys stored by customers were stolen in an October breach and used for an estimated $740,000 in unauthorized usage, potentially affecting up to 100,000 workspaces. The company promised to cover charges that AI providers will not refund.
Why it mattersStored AI API keys are now a direct target for theft and resale of compute, so set spending limits and rotate keys kept on third-party platforms.
Fireworks AI said its investigation into an earlier incident found further malicious activity affecting account and billing information, secrets, datasets and incomplete models for a limited set of accounts. Notion said the breach exposed Notion AI chat transcripts in a small number of workspaces.
Why it mattersYour AI data is only as safe as the infrastructure providers behind your tools, so companies should review which vendors process their prompts and data.
Microsoft researchers described a technique called Lazarus that changes a few dormant neurons in an open-weight model so that a trigger makes it output attacker commands, with no retraining needed. The modified model behaves normally in standard safety tests, and Microsoft says it has not seen attackers use the method.
Why it mattersTeams running agents on open-weight models need to verify where their model files come from and limit what agents can execute.
GitGuardian found 31,088 distinct secrets, including database credentials and cloud tokens, in Claude Code local settings files across 23,276 public GitHub repositories. Of those, 5,343 were still valid when detected.
Why it mattersIf you use Claude Code, check that your local settings file is excluded from Git and rotate any credentials that may have been committed.
A conversation with Sam Nickless, Partner and Chief Executive Officer at Gilbert + Tobin.
Google DeepMind's Gemini Decision 1 model evaluates up to 1,000 closed-ended questions at once across text, images, audio and video. Google says it returns results in 150 to 250 milliseconds at about five times lower cost than Gemini 3.5 Flash-Lite.
Why it mattersDevelopers doing classification, moderation or scoring at scale get a faster, cheaper option than general chat models.
The Office for Students says AI has made fair assessment of coursework nearly impossible after finding students submitting ChatGPT-written papers undetected. Universities must return to traditional exam formats or face sanctions.
Why it mattersIt signals a shift away from take-home coursework, affecting students, teachers and how degrees are earned.
No stories for this audience today. Try another filter.
Large frontier AI developers must publish safety frameworks and report safety incidents, with fines of up to $1 million for a first violation and $3 million for later ones.
United States (New York) · From 01 Jan 2027After a second delay, the law now mainly requires disclosures about automated decision tools and gives people limited rights to correct data and get human review of adverse decisions.
United States (Colorado) · From 01 Jan 2027The adopted Digital Omnibus (Regulation (EU) 2026/1744) pushed obligations for high-risk AI in areas like hiring, credit scoring, education and biometrics from August 2026 to December 2027.
European Union · From 02 Dec 2027AI built into products already covered by EU safety laws, such as machinery or medical devices, must meet high-risk requirements from this later date set by the Digital Omnibus.
European Union · From 02 Aug 2028
