Anthropic has been touting the facility of its Claude Mythos AI fashion to search out and assist repair safety vulnerabilities, and it has in any case launched a public model, Myth 5. Then again, some cybersecurity mavens are complaining that the guardrails supposed to forestall abuse are stifling its real-world usefulness.
Myth is billed as a “Mythos-class” fashion that may habits reasoning, coding, and cybersecurity duties higher than each the Mythos preview unencumber and standard AI fashions, together with OpenAI’s GPT 5.5 and Google’s Gemini 3.1 Professional. This preferably is helping safety groups pinpoint flaws, find out about problems, and expand fixes with relative velocity.
The variations between Myth 5 and its Mythos 5 identical come right down to the “safeguards,” in keeping with Anthropic. If the Myth fashion believes there is a chance of abuse, reminiscent of writing malware, it reverts to Claude Opus 4.8. Mythos 5 does not have as many protections, and get right of entry to is proscribed to participants of the Challenge Glasswing initiative supposed to safe important tool with the assistance of the U.S. govt and sure firms, together with Apple, Google, Microsoft, and loads of others.
The brand new Myth and Mythos variations are to be had at respective costs of $10 and $50 in keeping with million enter tokens, or not up to part the cost of the Mythos preview. Professionals who need fewer limits on any of Anthropic’s fashions must get approval via a Cyber Verification Program.
Why is Myth 5 so unhealthy for some cybersecurity mavens?
They really feel it is just too secure
Anthropic is obvious that it set Myth’s guardrails “conservatively” for the sakes of each a handy guide a rough release and protection, and that there can be moments the place “innocuous requests” prohibit capability. It expects to make stronger those protections and prohibit errors going ahead.
As TechCrunch explains, then again, cybersecurity veterans nonetheless don’t seem to be overjoyed. Professionals at IBM and in other places have identified that asking to write down or evaluation safe code downgrades to Opus, and that the boundaries would possibly kick in just by the usage of the ‘incorrect’ key phrase. For some, it may well be tricky to make use of Myth how it was once supposed.
5 AI Applied sciences Hackers Can Use in Horrible New Tactics
Some other people suppose The Matrix is an instruction guide.
The preliminary warning may well be warranted. AI can uncover and use vulnerabilities extra simply and briefly than sooner than, and that might result in zero-day exploits that builders have not had time to mend and even analyze.
Microsoft, for example, marked a file “Patch Tuesday” in June the place it mounted virtually 200 safety insects as its workforce and the neighborhood used AI to catch problems they would another way pass over. If attackers may just use gear like Myth or Mythos with out limits, there is a chance they may mechanically compromise methods sooner than patches are in a position, and with fairly little effort.
