Security
AI flaw-finder still under lock and key for now while company figures out guardrails, but extends access to more users including governments
Anthropic has revealed its intention to one day release models that match the performance of its Mythos bug-finding AI to the public, once it can make them safe.
In case you came in late, in early April Anthropic announced it had developed a model called Mythos that is so good at finding security vulnerabilities in programming code that the company decided to offer it only to select entities because allowing unfettered access would mean cybercriminals could quickly discover and exploit software flaws.
That access program is called “Project Glasswing” and participants report it quickly finds many bugs but few that humans couldn’t find given enough time and resources. Those with access to Mythos have also sometimes said the quantity of bugs it finds somewhat overwhelms their ability to patch them all.
The mere existence of Mythos has sparked a little panic – Japan’s government ordered a sweeping security review and Indian authorities demanded a patching spree at financial institutions – plus a general realization that even lesser AI models are also decent bug-finders, meaning cyber-defenders must now expect attackers will weaponize more flaws, more often.
No company—including Anthropic—has developed safeguards strong enough to prevent such models from being misused
Anthropic last week published an “initial update” on Project Glasswing that in its second-to-last paragraph reveals the company’s next step will see it “… work with critical partners – including US and allied governments – to expand Project Glasswing to additional partners. And in the near future, once we’ve developed the far stronger safeguards we need, we look forward to making Mythos-class models available through a general release.”
The company didn’t explain what it means by “near future” and admits that “At present, no company—including Anthropic—has developed safeguards strong enough to prevent such models from being misused and potentially causing severe harm.”
Further illustration of that assertion can be found earlier in the company’s post, which reveals that Anthropic has used Mythos to scan more than 1,000 open-source projects that it says “collectively underpin much of the internet – and much of our own infrastructure.”
To date, Mythos has found an estimated 6,202 high-or-critical-severity vulnerabilities in these projects – and 23,019 flaws in all.
The post reveals that when Mythos finds a flaw, Anthropic and its pals in the security community reproduce the issue that Mythos has found and “re-assess its severity.”
“Once we’ve confirmed that a vulnerability is real, we check for whether there are already fixes in place, and write a detailed report to the software’s maintainers,” Anthropic explains. “We take considerable care here: on top of the regular challenges of maintaining open-source software, maintainers have been facing a deluge of low-quality, AI-generated bug reports. Indeed, several maintainers have told us they’re currently severely capacity constrained, and some have even asked us to slow down our rate of disclosures because they need more time to design patches.”
1,752 of the high-or-critical-rated vulnerabilities Mythos found in FOSS have gone through that process and 90.6 percent (1,587) proved to be valid flaws. Of those, 62.4 percent (1,094) “were confirmed as either high-or-critical-severity,” the post states.
One of the critical flaws impacted the wolfSSL cryptography library used by billions of devices worldwide.
“Mythos Preview constructed an exploit that would let an attacker forge certificates that would (for instance) allow them to host a fake website for a bank or email provider,” Anthropic wrote. “The website would look perfectly legitimate to an end user, despite being controlled by the attacker.” Thankfully, developers have already patched wolfSSL, and Anthropic said it will deliver a full technical analysis “in the coming weeks.”
Keep an eye out for CVE-2026-5194 to learn more about this one.
Mythos is adding to an already overloaded security ecosystem
“75 of the 530 high-or-critical-severity bugs we’ve reported have now been patched, and 65 of those have been given public advisories,” the post states, then explains that low fix rate by revealing Anthropic is “still early in the 90-day window that’s set out in our Coordinated Vulnerability Disclosure policy: we expect many more patches to land soon.” The company thinks it is also “likely to be undercounting patches because some vulnerabilities are patched without a public advisory.” Lastly, the flood of bugs Mythos found “is adding to an already overloaded security ecosystem.”
Anthropic’s suggestion for security teams struggling to develop fixes for bugs AI discovered is, unsurprisingly, more AI such as skills that improve its Claude model’s ability to help developers. ®
