Anthropic Turning on FOSS Ecosystem?
In AI news, Anthropic says its new top-tier model, Claude Mythos Preview, will not be generally released. Instead, access is being gated through Project Glasswing: an invitation-only defensive-security research program centered on major firms, but also including the Linux Foundation and a wider group of infrastructure organizations. Anthropic says Mythos can find and exploit serious vulnerabilities across major operating systems and browsers, and can even turn many known flaws into working exploits. If those claims are true, then the risks go beyond what bad actors might do. It is also what happens if elite closed-source tooling advances faster than the open and auditable world can keep up with. In the worst case, we could end up in a world where closed-source products become technically harder to compromise than the open alternatives that people can actually inspect, modify, and trust. That would only deepen centralization at exactly the moment we need more diffusion, democracy, and citizen-led solutions in the software world.
Meanwhile, South Africa has published its draft National AI Policy for public comment. Much of it reads as sensible and rights-aware. But one part deserves scrutiny: the draft says the Information Regulator should be strengthened to counter "misinformation, disinformation and other harmful online practices," and it also gestures toward AI content-moderation and curbing misinformation. Read narrowly, that could mean fraud, deepfakes, and preventing real harm. But read more broadly, it raises the risk of mission creep into the realm of speech governance, depending on how "harmful" and "misinformation" are defined and enforced later. Whether these concerns are real remains to be seen.
Further north, the French parliament is now considering a bill that would shift the burden of proof in some AI copyright cases. Supporters say this is a necessary response to a lack of opacity in the AI industry, and thus gives writers, musicians, artists, and publishers a fairer chance to enforce their rights. Critics argue that it flips the normal burden of proof, and will expand litigation pressure, and thus chill lawful experimentation, satire, parody and stylistic imitation.
On a final note, users of NVIDIA's OpenShell sandbox on Linux should check a newly opened security issue on NVIDIA's NemoClaw tracker. The report says OpenShell 0.0.26 may fail to enforce its Landlock filesystem policy on tested setups, leaving parts of the host filesystem exposed to tampering by the AI agent. Anyone relying on OpenShell for containment should review the tracker and verify whether Landlock is actually active on their own system.
This video is provided for general informational purposes only and should not be considered legal, security, or privacy advice. Hero to Zero News makes no representations or warranties regarding the accuracy, completeness, or reliability of the information presented. Any views or opinions expressed on this channel are solely those of the speaker.
