TL;DR: Research from Wiz reveals 65% of Forbes’ top 50 AI companies are leaking verified secrets on GitHub, including tokens, API keys, and sensitive credentials. These exposures were found buried in deleted forks, developer repositories, and gists—locations traditional scanners miss. Nearly half of affected companies either failed to receive disclosure notifications or did not respond.

AI companies continue to demonstrate significant cybersecurity vulnerabilities despite their technological sophistication. New research from Wiz examining the Forbes top 50 leading AI companies found nearly two-thirds had exposed verified secrets on GitHub, creating substantial security risks.

The leaked materials include authentication tokens, sensitive credentials, and API keys buried deep in places most researchers and scanners would never encounter. These exposures were discovered in deleted forks, individual developer repositories, and GitHub gists—highlighting how secrets can persist in unexpected locations even after apparent removal.

Advanced Detection Reveals Hidden Exposures

Wiz employed a ‘Depth, Perimeter, and Coverage’ framework to uncover these vulnerabilities, enabling access to sources beyond surface-level scanning. The perimeter aspect expanded discovery to contributors and organisation members, who often “inadvertently check company-related secrets into their own public repositories and gists.”

Coverage addressed secret types frequently missed by traditional scanners, including tokens for services like Tavily, Langchain, Cohere, and Pinecone. This comprehensive approach revealed exposures that would escape conventional security scanning tools focused solely on obvious locations.

Disclosure Challenges Compound Risk

When researchers attempted to disclose these leaks to affected companies, nearly half of notifications either failed to reach their targets, received no response due to lack of official notification channels, or saw companies fail to reply or address the issues. This response gap compounds the security risk, as unaddressed exposures remain exploitable indefinitely.

The disclosure difficulties highlight a broader governance problem. Companies operating at the forefront of AI technology lack basic security infrastructure for receiving vulnerability reports—a fundamental requirement for responsible security posture.

Recommended Countermeasures

Wiz recommends immediate deployment of secret scanning as non-negotiable defence regardless of organisation size. Companies should prioritise detection for their own secret types, noting that “too many shops leak their own API keys whilst ‘eating their dogfood.’” For organisations with proprietary secret formats, proactive engagement with vendors and the open-source community to add detection support is essential.

Finally, organisations must establish dedicated disclosure channels. Disclosure protocol provides crucial early warning for vulnerabilities and leaks, making these channels a vital information-sharing resource. Without clear disclosure pathways, even well-intentioned security researchers cannot effectively communicate discovered vulnerabilities.

Source: TechRadar