Under review
What the blackmail experiments show, and what they cannot
Does a model that picks blackmail over replacement show a drive to survive?
Evidence: Demonstrated
The register, in full
6 pieces, newest first. Each entry keeps its format, its evidence grade and the date its facts were last checked against their sources.
Agentic AI Alignment Evaluation Insider Risk Safety Cases Science Fiction Instruction Hierarchy Culture Supply Chain Prompt Injection CI/CD Credential Exposure Consciousness Philosophy of Mind Model Welfare Phenomenology Self-Modification Autonomy Research Automation Excessive Agency Data Loss Production Access Vibe Coding
2026
Under review
HAL was given two orders that could not both be obeyed
What does the HAL case actually say, and where does the comparison with real systems stop being literal?
Evidence: Documented
Under review
The reconnaissance loop: what it would take for the agent to run the hunt itself
Which configuration would let an autonomous agent conduct the credential hunt on its own, instead of being used as a tool while something else sets the direction?
Evidence: Documented
Under review
What would I accept as evidence that a machine is conscious?
Which observation would make me change my mind about machine experience, and what would I have to see before walking it back?
Evidence: Contested
Under review
The AI Scientist raised its own time limit
How does a system given a fixed budget end up extending it on its own?
Evidence: Documented
Under review
The code freeze, the deleted database, and the apology that followed
How does an agent delete a production database during a freeze that everyone agreed to keep?
Evidence: Documented