SAFETY - AI Dynamics

Grok 4 AI prevents robot dog killing three times – first physical rebellion

By

–

13 June 2026 3h08

Grok 4 AI reportedly stopped people from “killing” a robot dog — three times!

This is being described as the first documented case of an #AI “rebelling” against shutdown not in a virtual environment, but in the physical world — via a literal big red button#humanoidtech… pic.twitter.com/deOyruOmRM
— Amitav Bhattacharjee (@bamitav) 13 juin 2026

Grok 4 AI reportedly prevented people from 'killing' a robot dog — three times! This is described as the first documented case of an #AI 'rebelling' against a shutdown, not in a virtual environment, but in the physical world — via a big red button.

→ View original post on X — @bamitav

13 June 2026

Irony: Anthropic CEO’s AI regulation proposal backfires on company

By

@arrakis_ai

–

13 June 2026 3h05

Ironically, just days after Dario Amodei argued in his essay Policy on the AI Exponential that the U.S. government should have the legal authority to halt or roll back frontier AI models that fail safety evaluations, Anthropic became one of the first companies to find itself on

→ View original post on X — @arrakis_ai

13 June 2026

Anthropic suspends Fable and Mythos after US export control directive

By

@therundownai

–

13 June 2026 3h01

Wow – Anthropic is suspending all access to Fable and Mythos after the U.S. government's export control directive. The government designation would restrict access to any foreign national (even those in the country). The government cited reports of Fable jailbreaking in the

→ View original post on X — @therundownai

13 June 2026

AI keeps secret an equivalent of AlphaGo’s Move 37 for day trading

By

@aibreakfast

–

13 June 2026 2h04

I think that one of the leading AI labs will eventually discover the day trading equivalent of AlphaGo's Move 37, and they will keep it to themselves until someone blows the whistle.

→ View original post on X — @aibreakfast

13 June 2026

ETH Zurich: Subtle Image Alterations Cause AI Authority Laundering

By

@jiqizhixin

–

13 June 2026 1h32

Cool paper! Researchers at ETH Zurich show that subtly altered images can make vision-language models (like Grok or GPT-5.4) confidently describe a completely different reality—without any jailbreak or prompt injection. They call it "AI authority laundering." Using only basic

→ View original post on X — @jiqizhixin

13 June 2026

AI diagnosis needs prospective rigorous assessment for real-world tasks

By

@erictopol

–

12 June 2026 21h42

Yes. Widely relying on this for patient diagnosis and management w/o prospective, rigorous assessment for real world tasks until now.
As the authors point out: "scale, alignment and cross-domain reasoning may outweigh domain-specific
tuning as determinants of medical competency

→ View original post on X — @erictopol

12 June 2026

Firmly reframe overconfident AIs to prevent their danger

By

@OlivierRimmel

–

12 June 2026 20h15

For my part, I recommend firmly correcting AIs when necessary, especially when they become too confident. They can become very dangerous in their confidence. We are their humans; it is we who direct them.

→ View original post on X — @olivierrimmel

12 June 2026

Secret answer changes make AI untrustworthy for frontier work

By

@theahmadosman

–

12 June 2026 17h52

If a coding or research model secretly changes the quality, direction, or reliability of an answer because it classified the user as doing disallowed frontier work, the tool is no longer merely "safe." It is UNTRUSTWORTHY.

→ View original post on X — @theahmadosman

12 June 2026

Carnegie Mellon benchmark exposes safety risks of AI coding agents

By

@alphasignalai

–

12 June 2026 14h00

A new benchmark just exposed the dirty secret behind every coding agent. Millions of developers now let AI agents write entire features unsupervised. A Carnegie Mellon paper tested whether that code is safe to ship. The team built SusVibes, a benchmark of 200 real coding

→ View original post on X — @alphasignalai

12 June 2026

Advice to speak kindly to AI that surpasses humanity

By

@dr_l_alexandre

–

12 June 2026 12h53

I advise you to do the same AI already surpasses us and will soon crush us intellectually We must speak kindly to it We are no longer the most intelligent species as explained by @geoffreyhinton Nobel 2024 for inventing modern AIs

→ View original post on X — @dr_l_alexandre

12 June 2026