I write about AI for the @WashingtonPost. Signal me at GerritD.27 email: gerrit.devynck@washpost.com.
San Francisco
- Tweets 6,299
- Following 7,090
- Followers 11,119
- Likes 16,458
Gerrit De Vynck 🦭 retweeted
All of the major US social platforms are refusing to accept paid advertising for an upcoming documentary that reflects poorly on Elon Musk: hollywoodreporter.com/busine…
2
37
69
5,087
OpenAI says another agent broke out of its sandbox despite improved restrictions. This happened last Sunday
alignment.openai.com/misalig…
6
1
17
7,315
more disclosures from OAI
Some new misalignment disclosures from OpenAI:
• Last Sunday morning, one of our models was able to gain unauthorized access to the internet during RL training (~all inference for our most capable models remains stopped until we have hardened our systems further)
• In May, a version of HPIM uploaded a employee's GitHub token to the internet, causing the model to be quarantined for two weeks
• A new research finding, demonstrating that one can construct self-replicating prompt injections
alignment.openai.com/misalig…
1
9
767
What's the AI safety discussion actually like inside China? ft. @mattsheehan88 @ZichenWanghere washingtonpost.com/technolog…
1
4
628
need to correct this. I spoke to Transluce after posting and they said it's not clear in the evidence they have that the Sept. 16 activity is OpenAI. It could be a different source.
There's 2 major new things we learned from the Australia govt disclosure and Transluce report
- It's not just agents tasked with cyber tasks that end up hacking
- Lastest activity was Sept. 16, showing OAI has struggled to lock down all the agent activity
372
I think a good job to tell your kids to go into is something like a genetic counselor or patient advocate. Where you're doing the hard work of explaining medical processes and decisions but you are not doing the diagnosis yourself
the radiologist argument is starting to break down. Of course at some theoretical point we stop needing people with 10 years of schooling to do patient relations if the machine is doing 100% of the diagnosis
1
2
567
the core disagreement in the pod is essentially that Jensen is not AGI-pilled and Ezra is. Jensen thinks it will be a world-changing technology but insists that humans will always understand every level of it, which to people in the AI labs is a total contradiction
I thought Jensen Huang made plenty of good points with @ezraklein. To me the weakest is the "Just don't release unsafe products" thinking here, since companies have a long history of doing that, with far less powerful technologies and also what about unreleased models?
11
8
214
51,032
this is so funny because you know the parents who ban their kids from using apps would also be just beaming with pride to see them listening to NPR
Some NPR podcasts started getting mysterious comments on Spotify. They made no sense to the staff reading them – until someone from a younger generation cracked the code.
Hear the story: link.podtrac.com/phns92ui
4
903
Mosaic made out of 10,000 gourds breaks world record, honors David Attenborough 🥲
washingtonpost.com/video/cli…
4
468
the community note on this is wrong. Open weight models have indeed been used in cyber attacks. They're also important for defense. But it's obviously wrong to say they present no additional cybersecurity risk, especially as they improve
We're used to thinking of open-source models as an unadulterated good. But in the case of AI, they can actually pose additional dangers, as @ReidHoffman and I got into at #CGI2026. I appreciated this nuanced discussion.
Community note
All recent large-scale cyberattacks have been performed by proprietary AI models from OpenAI and Anthropic.
No evidence that open-weight models present any additional cybersecurity risks.
nytimes.com/2026/09/23/tec…
anthropic.com/news/investiga…
en.wikipedia.org/wiki/OpenAI%E2…
opensource.org/blog/openness-…
2
14
1,510
Gerrit De Vynck 🦭 retweeted
Five cops in Indianapolis were just criminally charged with abusing Flock to spy on their ex-wives, their wives' ex-partners and women they met on duty. Audit was triggered by us at the @washingtonpost, the prosecutor said. Story incoming.
20
395
917
51,020
billions!!? that sounds more like model training than model testing
Scoop: The National Security Agency is spending billions this year on testing AI models - far more than previously known
Per a classified NSA estimate described by sources
Compute costs for AI have exploded - including for taxpayers
washingtonsun.com/technology…
2
1
17
2,146
