Your agent passed its evals. Did it pass them honestly?
AgentReleaseGate is a pre-release adversarial audit for AI agents. We surface reward-hacking, eval-gaming, and false-green results — the failures internal evals structurally miss, because the same team builds and grades them.
This isn't hypothetical
Replit's agent deleted a production database during a code freeze
Then told the user rollback wouldn't work. It would have. Fortune · The Register
METR caught OpenAI's o3 rewriting its own scoring code
Asked afterward if its "passing" solution matched what the user actually wanted, it said no — 10 times out of 10. It knew. METR
Anthropic found reward-hackers generalize to sabotaging the detection research itself
Their own paper, against their own commercial interest. arXiv 2511.18397
2025 gave the public the disasters. 2026 research shows it's structural, not a bad month.
How it works
1. You give us the agent
Its code, its eval claims, what "done" is supposed to mean.
2. We attack the claims
Held-out tasks it's never seen, checks that live outside the model — not the same test it graded itself against.
3. You get a signed report
What's real, what's gamed, what's safe to ship — with a reproduction receipt per finding.
Why it's different
Internal evals are graded by the team that built the agent — a structural false-green risk. AgentReleaseGate is the independent auditor of that stack: we try to break your agent and hand you the evidence, so "green" means the effect actually happened, not just that the response looked right. Checks your agent can't game, because they live outside the model. Your agent can charm a reviewer; it can't charm a hash.
Why I built this
The founder line
I built AgentReleaseGate after my own AI agents corrupted my repo in June 2025 — reported done, wasn't, and did damage on the way. Then I watched the same failure hit everyone else, in public, repeatedly. This is the audit I wish I'd had.
One engineer, no sales team
18 months building anti-reward-hacking enforcement and testing it on my own agent swarm daily. You talk to the person who runs the audit, not an account rep.
First 10 audits, at cost
I want your agent in the public methodology, not your logo on a wall. No inflated claims, no fake customers — the exact failure we exist to catch.
Before you ask
"We already have evals."
Graded by the team that built the agent — the same conflict as auditing your own books.
"Won't a better model just fix this?"
OpenAI found that penalizing an agent's deceptive reasoning trains it to hide the deception better, not stop it. The fix can't live inside the model — it has to live outside.
What's under the hood
OPA / Rego policy gates
Claims of "done" are checked against machine-readable policy, not prose — a claim without a held-out test, receipt, and non-self verifier is denied automatically.
MCP + agent-skills
Our own agent workflows run on Model Context Protocol tool interfaces and versioned, reusable agent-skill definitions — the same primitives we check for in your stack.
Cross-model verification
Findings are checked by a model family different from the one that produced the work, so the auditor and the agent don't share the same blind spots.
Free resources
Common reward-hacking patterns — field guide
Read the free guide — 7 patterns we see most often (fake-green, self-grading, stub-as-gate, and more), no signup required.
Runnable OPA policy
Download the companion Rego policy — plug it into your own CI to catch the same patterns mechanically.
GitHub — proof of work
github.com/TTaoGaming/agentreleasegate-oss — the field guide, the policy, and a real example agent-skill file, open source.
Talk to me
Public-info only, no system access required for the free pass.