Anthropic has confirmed that it plans to make its Mythos class of fashions publicly obtainable after delaying it resulting from safety dangers to private and non-private software program.
Mythos was introduced in April as a restricted mannequin, obtainable solely to pick firms reminiscent of safety researchers.
On the time, Anthropic cited the main “safety” dangers of the Mythos mannequin and determined to not launch it to the general public.
When Anthropic introduced the Mythos mannequin in April, it warned that “the benefit goes to those that can profit from these instruments.”
“If Frontier Labs is not cautious about how they launch these fashions, within the quick time period it may very well be the attackers. In the long run, we count on it to be the defenders who direct assets extra effectively and use these fashions to repair bugs earlier than new code is launched.”
AI firms usually keep away from deploying highly effective fashions till they develop robust guardrails to stop abuse.
Anthropic seems to have efficiently developed robust guardrails to stop abuse of the Mythos mannequin. The Mythos mannequin is taken into account to be rather more highly effective than Opus 4.8 and different fashions obtainable on the web.
Anthropic prepares to deploy Mythos mannequin
Anthropic confirmed in a weblog publish that it plans to launch Mythos class fashions to the general public within the coming weeks, however didn’t decide to a selected date.
“We’re shifting shortly to develop these security measures and count on to have Mythos-class fashions obtainable to all clients within the coming weeks,” Anthropic mentioned in a weblog publish.
Anthropic says it’s already permitting a small variety of organizations to make use of the Claude Mythos preview for cybersecurity work, but it surely’s unclear whether or not the identical mannequin can be made obtainable to the general public.
Based on the corporate, the Mythos mannequin affords important enhancements in code reasoning and autonomy, far exceeding Claude’s present flagship mannequin, Opus 4.8.
It is also value noting that the “Mythos-preview” mannequin was briefly seen to some Claude Code customers earlier than being taken offline.
Automated penetration testing instruments supply actual worth, however they have been constructed to reply one query: Can an attacker get by way of your community? They aren’t constructed to check whether or not controls block threats, detection guidelines fireplace, or cloud configurations are preserved.
This information describes six surfaces that you need to really study.
Obtain now
