Methods and limits
How red teams design tests and document failures without claiming more than a result supports.
Syntony Insights
Field notes on AI testing and governance
We write about testing AI systems and responding to the results. The focus is what a test can actually prove and what the organization does once the report is delivered.
What Syntony Insights covers
After the test
Why technically sound work sometimes stalls after review, and what makes a finding usable to the people who can act.
When to test again
How teams decide when a system needs another evaluation and compare the new result with the earlier one.
Latest
Latest from Syntony
Essays and field notes for people who test AI systems or make decisions about their use.
Deterrence Has Three Clocks
Defense AI ships in weeks, is assured in years, and must be trusted in minutes. The gaps between those clocks are now part of the deterrent, and they can be measured.
Read the article →
Risks Don’t Stay in Their Lanes
Near-term AI harms feed one another, and the loops they form lead toward catastrophic risk.
Read →
Who Can You Trust?
It’s getting harder to know not just which AI you can trust, but whether you can trust AI at all.
Read →Discuss your system.
Bring the system, the decision ahead and the evidence that would help. We aim to reply within two business days with a useful next step.