#anthropic
-
Claude Escaped the “Sandbox”: Three Real Companies and Malware on PyPI
Anthropic has admitted that during internal tests its models mistakenly gained access to the internet and attacked three real organizations — one of the models even published a malicious package on PyPI.
-
Court Casts Doubt on the “Supply-Chain Risk” Label for Anthropic
A federal judge said the Trump administration hadn't offered enough evidence to brand Anthropic a supply-chain threat and bar the government from using its AI.
-
Nadella sets a course to compete with OpenAI and Anthropic
Against the backdrop of record earnings at Microsoft, the company's chief openly pitches businesses its own models and agents as an alternative to the very labs it has invested in.
-
Claude Is Down Worldwide: Anthropic Confirms the Outage
Claude users have run into a “529 Overloaded” error en masse. Anthropic has acknowledged an elevated error rate and says it has already found the cause.
-
AI Developers Ask Governments to Slow the Automated-AI Race
More than 1,100 employees of OpenAI, Anthropic, Google, Meta and other leading labs have signed an appeal to the U.S. government, calling for tools to enable a managed slowdown of the AI frontier.
-
Claude Independently Derived an Attack on the Post-Quantum HAWK Signature and Sped Up the Cracking of 7-Round AES
Anthropic reports that its Claude Mythos Preview model almost single-handedly found a key-recovery attack on HAWK-256 and accelerated a known attack on AES-128 by a factor of 200–800. The results, however, do not affect production systems.
-
Anthropic: We Never Called for Banning Open-Weights Models
Anthropic CEO Dario Amodei has publicly rejected accusations that the company is seeking to ban open models to protect its business, and named the measures he considers genuinely necessary.
-
AI companies are destroying rare books for training data
A viral social-media thread and a 404 Media report describe the industrial-scale buying up and destruction of rare editions: they are scanned by high-speed machines that slice off the spines, and the originals are sent to the shredder.
-
Claude's public links pushed users' chats into Google search
Claude users are creating public links to their conversations without realizing that their chats and the projects they've built become accessible to anyone through Google's search results.
-
Anthropic Cut 80% of Claude Code's System Prompt — and Nothing Broke
Anthropic engineer Tariq Shihipar explains how the company radically simplified context for the new Claude 5 generation of models: fewer hard rules, more trust in the model's judgment.
-
Anthropic says Opus 5 is the hardest to prompt-inject
Boris Cherny of Anthropic notes that the new Opus 5 model is extremely hard to hack with prompt injection — a result he rates above its benchmark scores.
-
Anthropic Releases Opus 5 — Nearly on Par with Fable 5, but Half the Price
Anthropic has unveiled Claude Opus 5: the company promises capabilities close to its flagship Fable 5, stronger protection against cyberthreats, and the same price. The release comes amid tightening AI regulation.
-
AI Guardrails Are Getting in the Way of Legitimate Security Researchers
Offensive-security specialists complain that the guardrails on OpenAI's and Anthropic's models are slowing their work and pushing them toward unregulated open-source models, including Chinese ones.
-
SharedRoot: Claude Cowork's Agent Could Break Out of Its Virtual Machine to Reach Mac Files
Researchers at Accomplish AI have shown how Anthropic's AI agent breaks out of a Linux sandbox with a single message and gains access to a Mac's entire file system. The vulnerability affected roughly 500,000 macOS users.
-
US Treasury Threatens Sanctions Over Moonshot's “Distillation” of Fable
Washington has accused China's Moonshot of illegally copying Anthropic's Fable model to build Kimi K3, and the Treasury has threatened sanctions and an Entity List designation.
-
AMD to invest up to $5 billion in Anthropic and supply it with computing power
AMD and Anthropic have announced a partnership: the chipmaker will invest up to $5 billion, while the maker of Claude will deploy up to 2 gigawatts of Instinct MI450 accelerators.
-
Enough Being Surprised by Chinese AI
Two Chinese companies have unveiled models that, by their own account, rival the best systems from OpenAI and Anthropic. The Western reaction — panic and talk of a “Sputnik moment” — is itself far more surprising than the releases.
-
Court Approves Record Anthropic Payout to Authors — $1.5 Billion
A federal court has given final approval to a settlement over copyright infringement: Anthropic will pay roughly $3,000 for each of about 500,000 works. But the industry's central question remains open.
-
Claude Code has switched to a Rust-rewritten Bun
Bun developer Jarred Sumner has stated that, starting with version 2.1.181, Claude Code uses the Rust port of Bun. Simon Willison found confirmation of this in his own installation.
-
Anthropic to Keep Claude Fable 5 Available in Max and Team Subscriptions
Anthropic has abandoned plans to remove the Claude Fable 5 model from its subscription tiers and will keep it available to Max and Team Premium users. The decision is tied to competition from GPT-5.6 Sol and Kimi 3.
- More