A critical flaw in HEIF/HEIC/AVIF decoders allowed attackers to bypass many app defenses and remotely execute code or exfiltrate data across services, including OpenAI, Meta, GitHub Enterprise, and AWS, by exploiting image parsing tools.

Hackers exploited the libheif and libde265 vulnerabilities, in what researchers dubbed the HEIF Heist, to achieve remote code execution and data access across major platforms.

Questions arise about whether ecosystems will back-port fixes, label security updates, fund volunteer maintainers, or rely solely on upstream patches to stay safe.

The vulnerability was found in late July and patched within days, but deployments lacking the latest upstream fixes remain at risk.

CyberScoop is seeking comment from OpenAI about the findings and their broader implications.

Advances in AI may lower barriers to weaponizing quietly patched bugs, narrowing the window of safety created by patch delays.

Assigning CVEs and clearly marking security fixes is crucial for timely patching, as many tools and workflows depend on these signals.

Researchers used OpenAI and Anthropic AI systems to study the flaw, with Hacktron’s team aided by GPT-5.6 Sol and Opus 5.

Applying current patches is essential, as attackers could exploit similar decoders beyond OpenAI’s ecosystem if unpatched.

Because there was no CVE, many fixes didn’t trigger urgent updates, leaving vulnerable versions in widespread distributions including Debian base images.

Open-source maintainers, often volunteers, may not always classify fixes as security-relevant, revealing a systemic flaw in how security issues are prioritized.

OpenAI issued a $6,500 bug bounty for discovering the vulnerability, underscoring the broad reach given AI model integration across enterprise networks.