this took off. note that its not just an issue with copilot. see the responses. models use cat/grep to access env vars when they are literally told not to. antigravity runs commands that it feels is right without asking your permission.
SAFETY
-
Figure AI Sued Over Humanoid Robot Safety Allegations
By
–
4. Figure AI sued by ex-safety chief over alleged robot dangers A whistleblower claims Figure’s humanoid robots are powerful enough to “fracture a human skull” and cites a malfunction that gashed a steel refrigerator door. Figure denies all allegations.
-
Autopoietic ASI Agent V0 Superintelligence Breakthrough
By
–
[ V0.ASI.ETH ] Autopoietic ASI Agent (ASI.eth v0) #ASI #ASIFirst #Superintelligence
-
Agent Security Solution for AI Safety Implementation
By
–
Pretty common behavior; check out https://
github.com/mintmcp/agent-
security
… for a solution -
Anthropic Evaluates Honesty and Lie Detection in AI Models
By
–
8. Evaluating Honesty and Lie Detection in AI Models Anthropic researchers evaluate honesty and lie detection techniques across five testbed settings where models generate statements they believe to be false.
-
Data Leakage in Recommendation Systems from User Interaction Patterns
By
–
Good point. Btw, you also see a similar leakage trap in recommendation datasets. Most recsys data is built around users interacting with items over time, which means the same user appears dozens or hundreds of times. If you do a plain random split, you accidentally leak the
-
Why LLMs Struggle with Password Cracking Revealed
By
–
New insight into why LLMs are not great at cracking passwords! https://
share.google/j8kjt3yYLVfmhw
zMQ
… #LLM #LLMs #genai #GenerativeAI #AI #password -
We’re All Cooked: AGI Existential Risk May Be Blessing
By
–
we're all cooked, and that may be a blessing in disguise
-
AI 2027 Scenario: From Basic Agents to Superintelligence
By
–
From Agent-1 to Superintelligence: The AI 2027 Scenario The AI 2027 Report outlines one of the most thought-provoking trajectories for artificial intelligence: a rapid evolution from simple assistants (Agent-1) to autonomous, adversarially misaligned systems (Agent-4) and