CVE-2026-69110

The CVSS 9.1 score on this vulnerability is technically accurate but incomplete in a way that should concern defenders more than the raw number suggests. The 0.00549 EPSS reflects criminal incentive accurately — this is a self-hosted tool with limited mass-exploitation value — but it completely misses the actual threat profile in deployments where OpenCode Studio matters: strategic actors conducting targeted IP theft or infrastructure reconnaissance. What makes this more dangerous than EPSS implies: the vulnerability exposes intermediate artifacts from audio/video processing jobs — partial renders, encoding buffers, model inference traces, and pipeline configurations. These aren't final user outputs; they're the toxic waste of the AI processing pipeline. In many deployments, these artifacts contain system prompts, model version indicators, and processing parameters that constitute genuinely valuable intellectual property. An attacker who reads across multiple users' job artifacts can reconstruct your entire AI processing stack without ever exfiltrating a single user file. The DELETE endpoint compounds the risk in ways that should reframe your detection strategy. An attacker who can delete other users' processing artifacts doesn't just hide evidence — their destructive acts are indistinguishable from normal cleanup behavior. Job fails? User error. Artifact gone? Expected expiration. This creates operational camouflage that makes the attacker's presence nearly invisible in normal logs. The three affected endpoints share a root failure: static file serving that bypassed the authentication layer. This is not an isolated oversight. The pattern suggests authentication was either omitted or removed during performance work on async processing, and the gap was never audited because the endpoints felt like "infrastructure" rather than user-facing security boundaries. The critical defensive question is whether other undocumented processing endpoints exhibit the same failure. Prioritize: audit your temp directory contents immediately — assume any intermediate artifact may have been exposed and may contain pipeline intelligence. Implement detection for cross-user artifact access patterns, not just the volume-based alerts that work for opportunistic exploitation. The low EPSS is a signal about criminal opportunity, not a signal about actual risk in environments where this tool is deployed.

Reviewed through automated stages and approved by a human before publication.