No. 23Said
Within 3-12 months, unrestricted open models will match the best closed models at attack work, putting frontier capability in anyone's hands.
LinkedIn Have you ever thought about how easy it would be to hack you? (LinkedIn)
Unrestricted open source models are going to be as good as Sol / Mythos / Astra within 3-12 months, and that's basically going to be a Thanos Guantlet situation.
Daniel Miessler, 7 August 2026
First evidence: 1 month later
Aug 2026 → Sep 2026
- NIST CAISI: open-weight GLM-5.3 is the most cyber-capable open model to date but still about four months behind the US frontier: close, not equalnist.gov
- Anthropic: open-weight GLM-5.3 can build end-to-end cyber exploits as Mythos Preview can, though it is not shown to match the best closed models across the boardanthropic.com
He said it again
- We’re Getting the Wrong Message from Mythos
- X post: We’re missing a much bigger point on Mythos. It…
- X post: Mythos isn't even a cybersecurity model. It's just good…
- LinkedIn post: We’re missing a much bigger point on Mythos. It…
- Mythos is Just the New Normal
- X post: This whole Mythos thing continues to surprise me. Not…
- LinkedIn post: This whole Mythos thing continues to surprise me. Not…
- Unsupervised Learning NO. 526
- X post: What I don’t think people are understanding is that…
- LinkedIn post: What I don’t think people are understanding is that…
- X post: As someone who’s done offensive cyber for decades, I…
- Why We Need Some AI Model Controls
- LinkedIn post: My argument for why we need (some) limited controls…
- X post: Just wait til everyone has a Mythos-level Thanos Guantlet…
- How Easy It Would Be to Hack You
- Unsupervised Learning NO. 540
- LinkedIn post: unrestricted open models are about to put frontier-level attack…