Anthropic has not in fact released it, and it does in fact appear to be that dangerous, judging by the flood of vulnerability reports seen by e.g. Daniel Stenberg.
Certainly it’s a strategy OpenAI has used before, and when they did so it was a lie. Altman’s dishonesty does not mean it can never be true, however.
The flood of reports that open source projects like curl, Linux and Chromium are getting are presumably due to public models like Open 4.6 that released earlier this year, and not models with limited availability.
A few months of restricting access to people they think will actually fix problems is a big deal. Obviously only an idiot would think it could or should be kept under wraps forever.
> judging by the flood of vulnerability reports seen by e.g. Daniel Stenberg
Maybe I've missed anything, but what Stenberg been complaining about so far been the wave of sloppy reports, seemingly reported by/mainly by AIs. Has that ratio somehow changed recently to mainly be good reports with real vulnerabilities?
> Improvement in AI models' capabilities became noticeable early 2026, said Daniel Stenberg.
> He estimates that about 1 in 10 of the reports are security vulnerabilities, the rest are mostly real bugs. Just three months into 2026, the cURL team Stenberg leads has found and fixed more vulnerabilities than each of the previous two years.
> The challenge with AI in open source security has transitioned from an AI slop tsunami into more of a ... plain security report tsunami. Less slop but lots of reports. Many of them really good.
> I'm spending hours per day on this now. It's intense.
Partly true. I think the consensus was it wasn't comparable because Mythos swept the entire codebase and found the vulnerabilities, whereas the open models were told where to look for said vulnerabilities.
Not really. The models were pointed specifically at the location of the vulnerability and given some extra guidance. That's an easier problem than simply being pointed at the entire code base.
Surely the Anthropic model also only looked at one chunk of code at a time. Cannot fit the entire code base into context. So supplying an identical chunk size (per file, function, whatever) and seeing if the open source model can find anything seems fair. Deliberately prompting with the problem is not.
Certainly it’s a strategy OpenAI has used before, and when they did so it was a lie. Altman’s dishonesty does not mean it can never be true, however.