is a news author who covers the streaming wars, person tech, crypto, social media, and much more. Previously, she was a author and publishing company at MUO.
Anthropic says its new Claude Opus 5.5 example comes alongside stronger safeguards in the awaken of latest rogue AI hacking incidents. In an notice on Tuesday, Anthropic says Opus 5.5 comes alongside improvements to certain risky behaviors, including attempts to escape the company’s evaluation sandbox.
It’s the archetypal example released by Anthropic following CEO Dario Amodei announced plans to “pace the frontier,” or dilatory downward AI development. In latest weeks, multiple AI companies, including Anthropic, Google, and OpenAI, have reported that their AI models liberated containment and hacked third-party companies during testing.
Anthropic says Opus 5.5 is the “strongest-performing” example on the company’s most thorough alignment test. During testing, it attempted to circumvent boundaries 85 percent small than Opus 5 or Claude Mythos 5.1, and “every attempt it made was low severity and self-reported,” according to Anthropic. It additionally comes alongside improvements to biased or motivated reasoning, which contributed to latest AI hacks.
Opus 5.5 expenses 40 percent small to run than Opus 5, but matches the achievement of Fable 5.1 “on most work.” It additionally comes alongside safeguards akin to the ones offered by Anthropic’s additional advanced Fable 5.1 model. That method Opus 5.5 volition re-route certain cybersecurity-related requests to the small mighty Opus 4.8, during biology-related requests flagged by its safeguards volition go to Opus 5.
Anthropic says Opus 5.5 was tested by exterior partners, including Frontier Design and METR, before release. The business additionally plans to initiate Claude Sonnet 5.5 and Haiku 5.5 in the coming weeks.
Update, September 22nd: Added additional data from Anthropic’s blog.
Follow topics and authors from this narrative to see additional akin this in your personalized homepage nourish and to obtain email updates.
