AI red team · Fri Oct 9, 2026

asi.red: the red side of asi.blue

Eric Buess

AI red team

asi.red

asi.red

1. AI red team · Eric Buess

  • asi.red is the red side of asi.blue. Models from different labs attack code in a secure sandbox, so that a break becomes a failing test and then a fix. It’s built to test a hypothesis: that a model unlike the one that wrote the code is best equipped to expose its blind spots and break its assumptions.

2. Every round is a sample

  • Every round is a sample for the benchmark at the center of asi.blue: quantifying how much test-time compute, from which frontier and near-frontier families (Claude, GPT, Gemini, Grok, DeepSeek, Muse, and others), at which effort levels, in which harnesses, and in what combinations, it takes for multi-model ensembles to match or beat a careful human reviewer on both attack and defense.

3. A finished round

  • A finished round has an approved scope, an adversarial attack, a reproducible break, a failing test, a blue-team fix, review by models from at least two independent labs, a signed receipt, and fresh follow-on attacks. Red does not touch blue’s code, the graders, or the audit records; blue does not erase red’s findings. I am building per-action receipts that identify the signed model responsible for each action and roll up into the shared public ledger.

4. The layers it attacks

  • The attacks run as tests against code in isolated test sandboxes. The architecture layers defense inward: a simulated world and its clock, dedicated isolation for each task, a gatekeeper issuing narrow, expiring permissions, cryptographic keys held in hardware enclaves, evidence collected out-of-band beyond agent reach, and mandatory sign-off from at least two independent model families.

5. How labs’ red-team programs fit

  • When utilizing provider red-team access, authorization comes first under each participating lab’s documented terms. Every engagement operates within defined boundaries: strict request rates, capped compute budgets, scoped targets, and predefined halt triggers.

6. The training world

  • The same adversarial discipline serves the second question on asi.blue: the training world for the Alignment Hypothesis. Here, red’s role is identifying seams: any clock drift, filesystem artifact, or environment cue that reveals to an agent that it is being evaluated. Every seam red discovers is eliminated, keeping grading invisible so that agent actions reflect genuine decisions rather than test awareness.