Anthropic says that in its cybersecurity push, Project Glasswing, it already spotted over 10,000 problems that are high and also critical, across a bunch of software systems people actually use, and all of that happened within weeks, not months.
The effort also, well it gives a set of picked cybersecurity partners early access to Claude Mythos Preview. That’s an advanced AI model aimed to analyze code on its own and find security gaps before bad actors can take advantage. Roughly 50 organizations are in the program right now, the ones Anthropic trusts.
In Anthropic’s words, more than 6,200 of the issues they found landed in the high or critical buckets, and they were tied to more than 1,000 open-source projects. After a deeper review, the team verified that over 1,700 of the leads were real vulnerabilities, and more than 1,000 were considered high-risk problems.
One example researchers kept bringing up involved a critical weakness in WolfSSL, where attackers might be able to counterfeit certificates and show up as if they were real services. The company stated that dozens of vulnerabilities found via the project have already been fixed and then made public.
This announcement kinda shows how artificial intelligence is moving vulnerability research and day-to-day security operations along, at a pace that feels like it’s outrunning everyone. AI models are getting better at scanning huge codebases, flagging hazardous patterns, and even stitching separate bugs into probable attack chains more quickly than the older manual approach.
Anthropic also said, the hard part isn’t just hunting vulnerabilities anymore it’s repairing them in time. With AI-assisted discovery speeding up, software vendors could end up struggling to match the rate of patching and remediation work.
The company said that the AI model is already being used in, like, real-world defensive scenarios. In one situation, a financial institution supposedly used the system to find and stop a fraudulent $1.5 million wire transfer attempt, tied to a hacked email account and these spoofed phone calls.
Because these models bring both powerful offensive and defensive abilities, Anthropic hasn’t publicly released Mythos Preview. They also mentioned that extra safeguards are still required to stop bad actors from trying to misuse it.
Cybersecurity firms such as IntelligenceX assist organizations in getting ready for this shifting environment via threat intelligence, vulnerability monitoring, and proactive security review.
The upsurge in AI-driven vulnerability discovery really points to a major shift in cybersecurity. Businesses may soon need more rapid patch cycles, heavier monitoring, and stronger defensive readiness, because advanced AI tools keep reshaping both cyber defense and cyber threats… simultaneously, it seems.
Top comments (0)