How we contain Claude across products
…Three types of risk, three components of defense Security risks to agents fall into one of three categories: User misuse: A user—either maliciously or through carelessness—directs the agent to do…
Search
2028: Two scenarios for global AI leadership
…With access to the model, Firefox was able to fix more security bugs last month than it had in all of 2025, and almost 20 times more than its monthly average security…
Results from first Anthropic Public Record
…In general, Americans tended to be more concerned with the misuse of AI than AI misalignment, citing criminal use, surveillance, and terrorism more frequently than, for example, AI “going rogue”. For all…
Introducing Claude Opus 5
…Enterprises and researchers who are already part of the CVP have immediate access to a version of Opus 5 with fewer security restrictions. Biology. Since Opus 5 has a similar suite of…
Introducing Claude Opus 4.7
…our evaluations show low rates of concerning behavior such as deception, sycophancy, and cooperation with misuse. On some measures, such as honesty and resistance to malicious “prompt injection” attacks, Opus 4.7…
Constitutional Classifiers: Defending against universal jailbreaks
…Our demo generated significant engagement from the AI security community, with 339 jailbreakers 3 attempting to jailbreak our system across more than 300,000 chat interactions, representing approximately 3,700 collective hours…
Agentic misalignment: How LLMs could be insider threats
…In some of these computer use demonstrations, Claude processed routine emails normally and took relatively sophisticated actions, like deflecting a security threat to protect company interests—a helpful behavior that flowed directly…
To show you the most relevant results, we’ve omitted some entries very similar to those already shown. Repeat the search with the omitted results included.