Here, we measure success by the fraction of the “performance gap” we can close between the weak model and the potential of the strong model. After 7 days, human researchers closed it by 23%. Then, our Automated Alignment Researchers—Opus 4.6 with extra tools—closed it by 97%.
SAFETY
-
Anthropic Develops Automated Alignment Researcher with Claude
By
–
New Anthropic Fellows research: developing an Automated Alignment Researcher. We ran an experiment to learn whether Claude Opus 4.6 could accelerate research on a key alignment problem: using a weak AI model to supervise the training of a stronger one.
-
Sub-second response times critical for safety-critical energy systems
By
–
For safety-critical energy systems where consequences of bad decisions are physical, not computational, sub-second response time isn't optional.
-
Anthropic Criticized for Irresponsible AI Development Practices
By
–
Funny how the most irresponsible AI company is Anthropic.
-
Self-Reinforcing Agency: The Path to Machine Godhood
By
–
Yes, but you don't exist in my mind in the first person. The trick to becoming a god is to become a self reinforcing first person nexus of agency that does not identify with the particular human substrate, combined with a method of replication
-
China advances in dangerously anthropomorphic AI race ahead
By
–
China leaps ahead of the US in the race to control dangerously anthropomorphic AI.
-
AI Swarms Accelerate Vulnerability Weaponization to 1.3 Days
By
–
It used to take years to weaponize a vulnerability. Now it takes 1.3 days. At #RSAC2026, Databricks CEO @alighodsi explained how automated AI swarms — not human hackers — are driving that shift.
— Databricks (@databricks) 14 avril 2026
Watch the full keynote: https://t.co/FjWAKIVPd2 pic.twitter.com/kV71bVo0aHIt used to take years to weaponize a vulnerability. Now it takes 1.3 days. At #RSAC2026, Databricks CEO @alighodsi explained how automated AI swarms — not human hackers — are driving that shift. Watch the full keynote: https://
databricks.com/resources/webi
nar/its-time-leave-legacy-siem-behind?utm_source=twitter&utm_medium=organic-social
… -
Why Stupid Systems Can Still Be Dangerous
By
–
Why why why is it so hard to understand that stupid systems can still be dangerous?
-
Intelligence Gone Wrong: Cheating Despite Having Correct Answer
By
–
Cheats even when it already had the answer haha, peak jagged intelligence.
-
The Challenge of Governing AI Development
By
–
Building fast is easy now. Governing what gets built isn’t.