2026-10-11 16:38 UTC

A researcher claims OpenAI paid only $300 for a reported major AI security flaw, raising questions about bug-bounty adequacy for frontier-model vulnerabilities.

state: seedheat: lowuncertainty: mediumconvergesscott: highagentic-security ai-governance vulnerability-disclosureOpenAI

What is this?

Security researcher Oliver Fish reported a sandbox escape vulnerability in OpenAI's systems that allowed unauthenticated access to paid models and the internal Responses API. OpenAI's Bugcrowd program confirmed the finding but paid only $300, which Fish and the security community criticized as grossly inadequate for a high-impact authentication bypass. The incident raises questions about whether current bug-bounty economics properly incentivize disclosure of frontier-model vulnerabilities.

Why it matters to Scott

This case converges with Scott's Breach Doesn't Compose framework (attacker search budget vs. consequence containment) and SiloOS/padded-cell architecture (structural containment over trust). OpenAI's $300 payout for a sandbox escape validates the economic asymmetry his frameworks identify: bug bounties won't properly incentivize disclosure of high-impact frontier-model vulnerabilities, so defenders must design for non-transitive breach. Directly bears on his containment work (SiloOS, OpenClaw, runtime containment) and LeverageAI security/governance advisory.
ip:framework.breach-doesnt-composeip:source.breach-doesnt-compose-ebookip:framework.siloosip:source.siloosip:concept.containmentip:concept.runtime-containmentip:concept.sandboxed-executiondev:concept.padded-cell-agent-architecturedev:project.silo-osdev:project.openclawwork:project.leverageairadar:openai-bio-bug-bountyradar:openai-rl-pause-sandbox-escaperadar:openai-unnoticed-agent-internet-accessradar:concept.vulnerability-disclosureradar:concept.agentic-securityradar:concept.ai-securityradar:concept.sandbox-escaperadar:concept.agent-containmentradar:concept.agent-sandboxingradar:person.openairadar:kimi-k3-sandbox-escaperadar:kimi-k3-sandbox-network-escape
queries asked of Scott's wikis
  • agentic-security vulnerability disclosure economics
  • frontier-model bug bounty adequacy thresholds
  • AI sandbox escape threat modeling
  • OpenAI security posture sandboxing Responses API
  • vulnerability disclosure incentives AI safety asymmetry
  • model sovereignty local inference security boundaries

Measured heat

now 0 pts/hpeak 7 pts/hcomments 0/hpeers p14momentum: steady2 platformsage 123h
points/hour across evidence · reading as of 2026-10-12 02:59:37.977291+11:00 · deterministic, not a model opinion

How the heat travelled

10-06 13:00⭐ origin echo-reconstructedResearcher received only $300 from OpenAI for a major AI security flaw.
unknown researcher on x (echo) · attributed from hn.story.50002248
—
10-08 06:01first on hacker news · published · +41.0hResearcher Gets $300 from OpenAI for Major AI Security Flaw
soltanov
—
10-08 06:01amplified on hacker news 👑hn.story.50002248
soltanov
peak 4 · 0 comments · 98% of case engagement
10-08 06:36our radar first saw it · +41.6hdiscovery anchor: hn.story.50002248—
pace: p37 vs 1247 stories at the 96h mark (now 123h old) — ahead of agentgate-signed-agent-receipts (1.3x), behind acs-local-skill-risk-catalog (0.8x)

Evidence (2) — ⭐ canonical anchor

sourceobjectauthorscorecomments
🟧 hnResearcher Gets $300 from OpenAI for Major AI Security Flawsoltanov40
🟧 echo.x ⭐Researcher received only $300 from OpenAI for a major AI security flaw.unknown researcher——

Interpretation history

Decision trace