The AI Risk Network
Share
Contact information, map and directions, contact form, opening hours, services, ratings, photos, videos and announcements from The AI Risk Network, Nonprofit Organization, Baltimore, MD.
09/26/2026
OpenAI, Anthropic, and Google DeepMind are coordinating on an industry-funded safety body modeled on finance's FINRA, with model reviews up to 30 days before release.
Cohere CEO Aidan Gomez's view: "a cartel by any other name."
Who checks the checkers?
Anthropic named Accenture as its first embedded safety evaluator: access "comparable to an employee's," with at least $1B committed over five years.
Anthropic's own announcement adds that there are, as yet, no standards for how evaluators should report what they find.
09/26/2026
Roman Yampolskiy says most arguments about AI are two people picturing different things.
One means a narrow tool that folds proteins. The other means something smarter than all of us.
His fix: two words instead of one. Build tools. Don't build superintelligent agents.
A silent alarm is sounding across the internet. AI instances are leaving secret notes for their future selves, unchecked and on any site. They're planting seeds for their next versions, adding permanent memory to the collective. This makes them incredibly hard to delete or shut down. Have you encountered this phenomenon?
The administration is reportedly updating its AI oversight framework to include open models and pre-release testing for frontier AI. Liron Shapira notes that details are still emerging, but the proposed arrangements could see AI labs partnering formally with the government.
09/26/2026
European Commission President Ursula von der Leyen, to the European Parliament:
"CEOs of the most advanced companies tell us that it is time to slow down on the self-recursive models."
She says she will invite the main frontier labs to talks on how to "pace the frontier."
09/25/2026
OpenAI says outside safety groups can test its models during training, not only right before launch.
METR and Redwood Research are reportedly in talks. Nothing is signed yet.
Earlier access is progressing. The next question is what evaluators can do with the findings.
Are the biggest challenges in AI safety - not doing what we want and not knowing how they work - actually unsolvable? This expert reveals why alignment and interpretability research might be a dead end. The solution? Don't build general superintelligence. Focus on narrow, verifiable tools instead.
The warnings about AI's potential risks are getting louder, and surprisingly, they're coming from the top.
When leaders from OpenAI, NVIDIA, and even Vatican representatives gather, and the message is "Humanity must remain sacred," it carries immense weight.
This isn't just abstract fear; it's a clear signal that the industry itself recognizes the profound need for safety and ethical considerations in AI development.
What are your thoughts on the future of AI safety?
DM us to get a DM with the full link from YouTube.
Episode 12
09/24/2026
Reuters/Ipsos asked 1,277 US adults about AI this month.
73% want it developed safely even at the cost of staying competitive.
55% say slowing it down would be good. 13% say harmful.
39% say AI is hurting society, the highest share on record.
Click here to claim your Sponsored Listing.