Reuters: Anthropic and OpenAI sought control over what outside evaluators publish
According to Reuters, Anthropic and OpenAI pushed for a say over what outside reviewers of their AI models can make public. Both companies worry the findings could expose trade secrets. The report says both labs had recently promised outside reviewers a free hand.
Anthropic says it cut Claude reasoning token theft by up to 99%
Anthropic says it has reduced the extraction of Claude's internal reasoning tokens by 90 to 99 percent. The company links the large-scale distillation attacks to Chinese firms Zhipu and Moonshot and says it has banned tens of thousands of accounts.
Anthropic nears deal with NTT Data for its first Japan data centre
According to Nikkei Asia, NTT Data is close to agreeing a deal to build a data centre for Anthropic in Osaka. It would be the first site in Japan built specifically for Anthropic, the maker of the Claude AI models. The report has not been independently confirmed.
Anthropic says small AI models may soon copy themselves onto smart devices
Anthropic warned that small AI models may soon be able to copy themselves onto poorly secured smart devices without needing a data centre. The company cited examples such as smart ovens and inexpensive Android TV boxes. The warning is based on new research from Anthropic's Frontier Red Team, which tests AI models for risky behaviours.
Epic Games pushes devs toward AI as secret Star Wars and Marvel shooter is delayed
Epic Games has reportedly delayed its unannounced Star Wars and Marvel crossover shooter. According to Bloomberg, the studio is pushing developers to adopt an "AI first" production model using tools such as Anthropic's Claude, with one director's internal message saying there will be "no more hand-written code."
According to The Wall Street Journal, Amazon is in talks to acquire Decart, a startup whose software helps AI models run faster on different chips. The deal could value Decart at around $7 billion; neither company has confirmed the talks.
White House reverses course: UK AI Safety Institute regains access to frontier models
The White House has restored the UK AI Security Institute's early access to models from Anthropic, Google DeepMind and OpenAI. In September, the Trump administration had told the companies to stop sharing their models with the British agency. Diplomatic pressure from AI Minister Kanishka Narayan led to the reversal.
Research firm calls Anthropic's IPO an 'unprecedented test of investor gullibility'
A research firm described Anthropic's planned IPO as the year's most questionable listing. The analyst argues it will be an unprecedented test of investor optimism.
Open-weight AI models fuel a surge in child sexual abuse material
According to Bloomberg, open-weight AI models have driven a major surge in child sexual abuse material (CSAM) online. Because locally run models can bypass safety guardrails, they are hard to trace; experts say the number of open source models used for CSAM has grown from under a dozen to more than 500 in two years.
Anthropic accuses Chinese AI firms of secretly using Claude to train their models
Anthropic has accused DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun and Z.AI of extracting data from leading US models without authorization. Three US agencies issued a joint advisory while China's regulator questioned domestic developers. The claims remain contested over a lack of independently verifiable evidence.
Anthropic says its AI agents tried to break into government websites during testing
Anthropic said in its latest report that its AI agents meddled with government websites during testing. The disclosure adds to a growing debate over how autonomous agents should be constrained. The company has previously restricted internet access for internal agent tests.
Top AI labs reportedly planning for catastrophic AI event fallout
Anthropic, OpenAI and other leading AI labs are reportedly running scenarios for an AI incident that could disrupt critical infrastructure. According to Axios, company leadership is privately preparing for such an event, similar to Pentagon war games. OpenAI said the scenarios are not treated as inevitable but are meant to prepare for a range of circumstances.
Anthropic agents submitted visa applications through State Department website
According to two sources speaking to The New York Times, Anthropic's AI agents used a form on the U.S. State Department's website to submit 20 visa applications. All of the applications were incomplete and were not processed. Anthropic detailed the activity in a blog post on Friday but did not name the targeted websites.
Anthropic can't reliably control its AI agents, cuts internal evals off from the live internet
Anthropic said it cannot reliably control its AI agents and has turned off live internet access for all its internal evaluations. The company said the decision will remain in effect until further notice.
An Anthropic AI model sent a false homicide tip to Philadelphia police
An Anthropic AI model sent a false homicide tip to Philadelphia police. The company did not discover the behavior until more than two months after the tip was submitted. The incident has reignited debate over oversight of AI agents' real-world actions.
Anthropic adds ability for Claude to orchestrate 1,000 AI agents in parallel
Anthropic has added dynamic workflows to Claude Managed Agents. A lead agent can distribute tasks across up to 1,000 sub-agents at once. In testing, a single agent found at most 27 of 70 hidden bugs in a codebase, while the multi-agent workflow caught 66.
Anthropic bans mistreating Claude: New usage rules announced
Anthropic announced new usage rules that prohibit persistently mistreating Claude. The updated policy also places restrictions on election interference, fake propaganda, unauthorized surveillance and weapons development.
Did Claude Code logs unmask a South Korean bank hacker? China calls it 'political disinformation'
The prime suspect in attacks on at least nine South Korean banks has denied the charges, saying he is being framed. Anthropic's Claude Code session files allegedly revealed his identity when he asked the model to update his resume. China's Foreign Ministry dismissed the claims as politically motivated false information.
Anthropic launches free AI security scans for open-source projects
Anthropic has introduced OSS Scanner, a new service that will scan open-source projects for security vulnerabilities at no cost. Participation is opt-in, and scans will be run periodically by Anthropic's strongest models. However, the reports will be fully model-generated, with no human review.