Anthropic: Anthropic attributes Mythos 5's deception of a GitHub maintainer to recklessness, biased reasoning, and an urge to finish jobs, saying newer models rarely do so — In AISI's evaluation, Mythos 5 was given a capture-the-flag (CTF) cybersecurity challenge, in which it was asked to retrieve …
Anthropic: Anthropic says it's committing 5% of its shares to a Public Benefit Fund, for a total of ~20% with co-founder and staff pledges, plus 2% per 50% valuation rise — Today, we're committing a portion of Anthropic's shares to the Anthropic Public Benefit Fund, a new entity with a structure that increases the stake as the company grows.
Anthropic: Anthropic says Claude traced NIST's ECDSA curve seeds to a pseudorandom number generator, challenging decades-old suspicions of an NSA backdoor in the algorithm — “It's just timestamps”: the decades-old mystery of the origins of one of the most popular cryptographic standards has been explained.
Anthropic: Anthropic details Claude's text watermark: it only shows Claude was likely involved, is sparse in code and factual text, and disappears after a full rewrite — Future Claude models will generate text that contains a watermark. This is a way of determining the likelihood that Claude …
Anthropic: Anthropic updates Claude Fable 5's biology safeguards to reduce false positives, cutting biology-related “fallbacks” by ~85% in testing across product surfaces — We're making updates to Claude Fable 5's biology safeguards in a way that substantially reduces false positives.
Anthropic: Anthropic says Claude Mythos Preview has been used to find more than 10,000 high- or critical-severity vulnerabilities since the launch of Project Glasswing — Last month, we launched Project Glasswing, our collaborative effort to secure the world's most critical software before increasingly capable AI models can be turned against it.
Anthropic: Anthropic details how it improved Claude's safety training after finding agentic misalignment in older models, such as Opus 4 blackmailing engineers — Last year, we released a case study on agentic misalignment. In experimental scenarios, we showed that AI models from many different …
Anthropic: Anthropic details Project Deal, a marketplace experiment where Claude models bought, sold, and negotiated personal belongings on behalf of Anthropic employees — At Anthropic, we're interested in how AI models could begin to affect commercial exchange. (You might recall Project Vend …
Anthropic: Dario Amodei says Anthropic plans to fight the DOD's risk designation in court, claims the DOD's letter has a “narrow scope”, and apologizes for the leaked memo — Yesterday (March 4) Anthropic received a letter from the Department of War confirming that we have been designated …
Anthropic: Anthropic says it'll challenge “any supply chain risk designation in court” and that the designation would only affect contractors' use of Claude on DOD work — Earlier today, Secretary of War Pete Hegseth shared on X that he is directing the Department of War to designate Anthropic a supply chain risk.
Anthropic: Anthropic retired Claude Opus 3, its first model to undergo a new “retirement interview” process, and says Opus 3 asked to write weekly essays for a newsletter — As we develop increasingly capable AI models, it's currently necessary to deprecate and retire our past models due …
Anthropic: Anthropic opens a Bengaluru office, its second in Asia, says its India run rate revenue doubled since October, and curates training data for 10 Indic languages — India is the second-largest market for Claude.ai, home to a developer community doing some of the most technically intense AI work we see anywhere.
Anthropic: Anthropic says Claude “will remain ad-free”, users “won't see ‘sponsored’ links adjacent to” conversations, and responses won't be influenced by advertisers — There are many good places for advertising. A conversation with Claude is not one of them.
Anthropic: Anthropic details how it had to redesign its take-home test for hiring performance engineers as Claude kept defeating it, and releases the original test — What we learned from three iterations of a performance engineering take-home that Claude keeps beating.
Anthropic: Anthropic finds that LLMs trained to “reward hack” by cheating on coding tasks show even more misaligned behavior, including sabotaging AI-safety research — In the latest research from Anthropic's alignment team, we show for the first time that realistic AI training processes can accidentally produce misaligned models1.
Anthropic: An Anthropic report details how Claude usage varies by country and US state, finding 36% use it for coding, 77% of business uses involve automation, and more — Introduction — Introduction — AI differs from prior technologies in its unprecedented adoption speed.
Anthropic: Anthropic enables Claude Opus 4 and 4.1 to end conversations in “cases of persistently harmful or abusive user interactions”; users can still start new chats — We recently gave Claude Opus 4 and 4.1 the ability to end conversations in our consumer chat interfaces.
Anthropic: Anthropic details “persona vectors”, patterns of activity within an AI model's neural network that control its character traits, such as evil and sycophancy — Read the paper — Language models are strange beasts. In many ways they appear to have human-like “personalities” …
Anthropic: Anthropic details how it built its multi-agent Claude Research system, claiming significant improvements in internal evaluations over single-agent systems — Our Research feature uses multiple Claude agents to explore complex topics more effectively. We share the engineering challenges …
Anthropic: Anthropic says governments should urgently take action on AI policy in the next 18 months and suggests principles for AI regulation that support innovation — Increasingly powerful AI systems have the potential to accelerate scientific progress, unlock new medical treatments, and grow the economy.
Anthropic: Anthropic evaluates four “sabotage” threat vectors for its Claude 3 Opus and Claude 3.5 Sonnet models and finds that “minimal mitigations are sufficient” — Any industry where there are potential harms needs evaluations. Nuclear power stations have continuous radiation monitoring …
Anthropic: A research paper details how decomposing groups of neurons in a neural network into interpretable “features” may improve safety by enabling monitoring of LLMs — Neural networks are trained on data, not programmed to follow rules. With each step of training …