The guidelines will establish federal safety and performance baselines.
The White House is accelerating plans to announce the new AI model standards and guidance, including the new benchmark co-designed with Anthropic, they could arrive as soon as next week according to the Financial Times.
Users doubt Anthropic co-designing the AI model standards benchmark fearing heavy-handed safety focus or weak voluntary rules, while some praise the administration's fast proactive collaborative approach.
No Digg Deeper questions have been answered for this story yet.
Financial Times:
- USG is in talks with AI companies (including at least Anthropic, OpenAI and Google) to create *voluntary* standards for release of new AI models, to be announced "as soon as next week".
- Standards will include: (i) benchmarks for models with "cutting-edge CYBER capabilities"; (ii) release timelines (which is important after Fable 5 and GPT-5.6 were lost in no-man's land for a while); and (iii) more clarity around the definition of a "frontier" model to which these rules will apply.
- CAISI(!) and the NSA will play a crucial role in setting and monitoring the standards.
- The USG will clarify who is able to access models, "both domestically and abroad, in a move that could set the stage for a global framework including US allies".
- Wider release of GPT-5.6 is expected "as soon as next week".
- Google has been in discussions with the USG about release about its own "advanced coding models", with "more sophisticated cyber capabilities than prior generations".
@AndrewCurran_ If Anthropic's helping design it, I'm not sure that's a good thing.
Financial Times:
- USG is in talks with AI companies (including at least Anthropic, OpenAI and Google) to create *voluntary* standards for release of new AI models, to be announced "as soon as next week".
- Standards will include: (i) benchmarks for models with "cutting-edge CYBER capabilities"; (ii) release timelines (which is important after Fable 5 and GPT-5.6 were lost in no-man's land for a while); and (iii) more clarity around the definition of a "frontier" model to which these rules will apply.
- CAISI(!) and the NSA will play a crucial role in setting and monitoring the standards.
- The USG will clarify who is able to access models, "both domestically and abroad, in a move that could set the stage for a global framework including US allies".
- Wider release of GPT-5.6 is expected "as soon as next week".
- Google has been in discussions with the USG about release about its own "advanced coding models", with "more sophisticated cyber capabilities than prior generations".
The White House is accelerating plans to announce the new AI model standards and guidance, including the new benchmark co-designed with Anthropic, they could arrive as soon as next week according to the Financial Times.
@AndrewCurran_ Huh? Even though Anthropic is a company that could potentially undermine national security, working with them to build safety together... isn't that a crisis for America?
@AndrewCurran_ I hope the benchmark is co-designed with not just Anthropic.
https://www.ft.com/content/0bb7e2f9-007b-4577-9c4a-858948ee969a?syn-25a6b1a6=1
@AndrewCurran_ Lost me at 'co-designed with Anthropic'. Ugh.
If there is one thing I've learned in my lifetime of business- you steer future strategy in your favor by making sure you are always involved with the decision making.
@deredleritt3r > Google has been in discussions with the USG about release about its own "advanced coding models"
@AndrewCurran_ @gladstein Really not a good idea to have anthropic designing this benchmark. Oh my God we're going to be consumed by safety.
@deredleritt3r That "voluntary" has the strength of a wooden roof through a hurricane.
@AndrewCurran_ With AI, every day is like the count in ZeroDark30 - we all need to keep tapping on the glass
@AndrewCurran_ Why are they not designing benchmarks in collaboration with all of the leading AI labs? This seems odd.
@AndrewCurran_ I just hope it allows for the rapid releases we started getting these last few months
That was a fun period, getting major updates closing in on once a month and promising even quicker as we continued to accelerate. That era ended far too early.
@deredleritt3r I would be absolutely shocked if Gemini 3.5 Pro turned out to have sophisticated cyber capabilities.
@AndrewCurran_ Hopefully this accelerates exports.
@AndrewCurran_ Things can feel slow during the singularity, but this might be the fastest-moving administration of all time.
@AndrewCurran_ This is not good news in the slightest.
@AndrewCurran_ If they designed with Anthropic it’s likely to be heavy handed .
@AndrewCurran_ I just hope the government isn't as ludicrously stupid as the reporting around fable 5 made them seem like "it was able to do a jailbreak called 'code review'" style reporting you'd think people can't be that stupid but then.. https://youtu.be/lTonHRerMC4
@AndrewCurran_ curious to see the benchmark