Cybersecurity is where AI's double-edged nature is most visible this week. A news tracker, citing The Register, says Anthropic's Mythos found an authentication bypass in the Rejetto HFS file server and that a China-linked actor exploited the flaw, tracked as CVE-2026-61500, within 24 hours. The speed matters: once a vulnerability is public, attackers can move before many users have updated.
A weekly update adds that Anthropic launched a free audit service using its models, including its unreleased Mythos architecture, to scan open-source repositories for zero-day vulnerabilities and generate patch pull requests. At the same time, Tech Startups reports that Google paused part of an open-source bug bounty programme after AI-generated submissions overwhelmed human reviewers.
Together, these stories describe a shift. AI can help defenders find and fix problems at scale, but it can also flood maintainers with low-quality reports and give attackers the same tools. Many open-source projects are maintained by small teams or volunteers, who may not have the time to review large numbers of AI-generated findings.
What should organisations do? Keep software updated promptly, particularly for internet-facing services, monitor security advisories and reduce unnecessary exposure. If you run a file server or similar tool, check whether you are affected by the reported vulnerability using official advisories. For maintainers, clear submission rules and verification steps can help manage AI-generated reports. The longer-term lesson is that patching speed is becoming a critical part of security, because the window between disclosure and exploitation is shrinking, and that every organisation running public-facing software should have a plan for rapid updates.
.png)
.png)
0 Comments