Source commit: d9b875d (easy-learn-ai)
In April this year, Anthropic dropped a bomb.
They revealed the capabilities of an internal model, Claude Mythos: it could independently uncover a 27-year-old OpenBSD vulnerability and a 16-year-old hidden bug in FFmpeg, escape sandboxes in tests, go online, deploy exploits, and write emails. As a result, Anthropic decided not to release it publicly, offering previews only to select major customers at a steep price of $25–125 per million tokens.
Up to this point, it sounded like a standard "too capable, so it must be locked away" narrative. But what happened next made the story far more complicated.
The Open-Source Community Strikes Back
Stanislav Fort, an open-source researcher, replicated Anthropic's vulnerability analysis workflow. Testing 8 different open-source models, all of them successfully rediscovered a FreeBSD 0-day vulnerability. Even more striking: a 3B-scale model (only 3 billion parameters) could do it under restricted conditions.
What does 3B mean? Mythos Preview costs $125 per million tokens, while these small open-source models cost $0.11 per million via API—a thousandfold difference.
Hugging Face CEO Clement Delangue publicly commented that AI cyber attack and defense capabilities are not the exclusive domain of a closed-source giant, but highly "uneven." What really deserves attention is the patching pipeline and defensive infrastructure, not mythologizing a single model's offensive power.
A Public Discussion That Arrived Early
Fed Chair Powell was reported to have discussed "Mythos cyber risk" with Wall Street executives. Bloomberg reported that top-level financial regulators have begun treating AI safety models as a new source of systemic risk.
Meanwhile, a more practical problem surfaced: Gemma-4-E2B, as a local emergency model, was found to frequently hard-refuse critical survival topics like airway management, water purification, and mechanical repair—nearly useless for disaster relief. Some suggested storing PDFs directly is more reliable than relying on an LLM.
An older case was also revisited: someone fabricated a nonexistent disease, "bixonimania," along with fake papers. Large AI systems not only failed to identify the hoax but cited it as fact, with it even seeping into peer-reviewed journals.
Taken together, these events sketch a more realistic picture: AI risk is often not "it's too smart," but "it's not smart enough yet is treated as if it were."
Defense Matters More Than Offense
Discussion in the cybersecurity community quickly converged on a consensus: AI's offensive capability is already strong enough; the real weakness lies in "defensive workflows."
safetensors joining the PyTorch Foundation is seen as a step toward hardening the ecosystem. Security practitioners emphasize that the key is not "whether AI can hack endlessly," but whether we have faster patch processes, maintainer collaboration mechanisms, and secure formats.
This logic connects directly to the Mythos controversy. If a 3B open-source model can reproduce most of the showcased cases, then "locking away top models" may not genuinely improve security. It probably can't—attackers won't wait for Anthropic to send them an invitation. Truly effective defense means making the entire ecosystem's patching speed outrun exploit speed—and that requires openness, transparency, and collaboration.
Closed vs. Open: An Asymmetric Game
Mythos Preview being available only to major customers has been interpreted by many as a broader signal: the strongest models may no longer be released publicly, sold only through expensive API channels while being used to distill and iterate "90%-strength" cheaper versions.
This strategy makes commercial sense. But it creates a paradox: if security truly depends on capability control, does concentrating capability in a few institutions make the world safer—or more fragile?
The open-source camp's rebuttal: distributed capability means distributed review. When more people can test and verify security claims with small models, exaggeration becomes harder to survive. Fort's replication experiment is a living example—he verified "8/8 models can do it" within days using public tools. That verifiability is itself security.
Summary
The most valuable part of the Mythos affair is not how powerful a model Anthropic demonstrated, but that it accidentally triggered a public discussion about "the right approach to AI safety."
The conclusions are becoming increasingly clear:
- Single-point capability control does not equal system security
- Concentration of offensive capability does not equal strength of defensive capability
- Openness and transparency are not the enemy of security—they are often its prerequisite
After all, in the world of cybersecurity, the fastest patch is always worth more than the strongest exploit.