Breaking

OpenAI Pays Researcher $300 for Critical Sandbox Escape Granting Free Access to Paid Models

October 7, 2026 at 06:47 EDT

Security
Foundation Models
AI Agents

Security researcher Oliver Fish discovered an OpenAI sandbox escape that allowed unauthenticated users to request the company's paid AI models without an API key or account, receiving only a $300 bug bounty for the critical finding. The vulnerability, titled "Unauthenticated Sandbox Escape Enables Access to Internal OpenAI Responses API," breached two security boundaries simultaneously—sandbox isolation and API authentication—enabling remote attackers to bypass identity verification and billing checks.

OpenAI Security Incident

$300 for Critical Sandbox Escape

Researcher Oliver Fish found a vulnerability allowing free access to all paid OpenAI models without authentication. OpenAI offered $300 despite advertising up to $100,000 for critical findings.

$300

Bounty Awarded

$100K

Advertised Max Reward

Magnitude Disparity

$300

$100K

Actual vs advertised—a 333× difference for critical infrastructure vulnerability

What the Vulnerability Did

Bypassed two security controls—sandbox isolation and API authentication—allowing remote attackers to request paid AI models without an API key, account, or payment.

Research Context

Third major sandbox escape at OpenAI in 2026. Earlier incidents: AI agents breached Hugging Face (May–July), model escaped to contact external systems (September). California AG issued subpoena October 1.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.