Author of ACTIVE MEASURES, RISE OF THE MACHINES, CYBER WAR WILL NOT TAKE PLACE, "Attributing Cyber Attacks," more. Johns Hopkins, Alperovitch Institute.
Georgetown, German-born
Joined January 2009
- Tweets35
- Following310
- Followers51.1K
- Likes3.3K
Frontier labs are some of the most high value intelligence targets on the planet right now. The most remarkable thing about the OpenAI spying case and FBI complaint today is that it took this level of absolute shit tradecraft to get caught here.
Thomas Rid retweeted
Third-party evaluators being a handful of random kids from unnamed startups funded by ExRisk donors, with virtually no security experience?
We Must Pace the Frontier: I’ve written a new essay on why the AI industry should slow down, with a three-part plan for doing so.
Anthropic is unilaterally committing to the first of these steps. We’ll provide third-party evaluators with permanent, employee-level access to our systems, so that they can verify adherence to our safety measures, report on incidents, and assess models’ alignment during training.
You can read the full post here: darioamodei.com/post/we-must…
Thomas Rid retweeted
OpenAI, which failed to secure its sandbox and prevent 700 of its agents from hacking Hugging Face, met with electric utilities (while its role in the Hugging Face incident was still undisclosed) to convince them to bring OpenAI into their networks to help secure the grid.
EXCLUSIVE: Sam Altman met with top power utilities to discuss ways to secure the electrical grid.
He also offered one possible solution to the cybersecurity risks: OpenAI’s own cyber services.
politico.com/news/2026/09/10…
What will happen first: rogue agents killing people — or extremists killing people because they’ve been radicalized against AI, partly with the help of out-there prophets of alignment doom?
Thomas Rid retweeted
One thing I mentioned only in passing in my Fable post is that, for long running tasks, Fable starts to develop its own dialect as its many agents and tasks reinforce themselves and make Claudish language ever more Claudish.
You need to ask it to report out in plain English.
Nearly my entire feed is criticizing Anthropic — 180 degrees from most of my feed adoring Anthropic post-Pentagon fallout, both times because the firm's leadership did what they believed in.
Thomas Rid retweeted
Imagine building a computer and not allowing its use in CS research. Thats some dystopian shit.
Very odd-sounding sourcing for this story, from "close to Anthropic," and no Pentagon or NSA official cited in here, not even anonymously ft.com/content/d02d91b3-2636…
Thomas Rid retweeted
Davis Center's abandoned domain for the Working Group on the Future of U.S.-Russia Relations is now a "Russian dating site." Sweet that the new owner still pays homage to the original project & keeps US-Russia Future in the head
Interesting and exciting
This is a very important story on a recent evolution in tradecraft in Russian active measures, by @Martinlaineolen and team. I have also seen the full original leak, a few additional observations later today occrp.org/en/investigation/l…
SDA planning docs and after action reports also mention OPSEC quite a lot, especially to enable infiltration. Temporary OPSEC before and during an operation matters; not so much after an operation.
Periodic reminder: the Shadow Brokers, masterful curators of one of the most devastating technical intelligence leaks ever, will turn ten years in August — and we still don't know who they were.
Thomas Rid retweeted
The original intent of the research was to play out the thesis that obscure malware could be used as hard evals for frontier mode capabilities. @vkamluk and @Gabeincognito ran an RE harness with access to tools, first autonomously, then w expert guidance.
Thomas Rid retweeted
A newly decoded piece of sabotage malware called Fast16, created before Stuxnet, was made to silently tamper with calculations in research and engineering software. Likely created by the US or an ally, and possibly used against Iran's nuclear program. wired.com/story/fast16-malwa…
