Cisco
Sources: Cisco (homepage) · OSS AI Security Harness Comparison.
Identity and role
Cisco is a networking and cybersecurity vendor. Its Cisco Open program publishes cisco-open/ai-deep-sast on GitHub, a SAST+LLM hybrid vulnerability scanner named in Semgrep’s July 2026 survey of open-source AI code-security harnesses.
Relevance to This Wiki
ai-deep-sast holds about 50 GitHub stars, the fewest of the nine projects in Semgrep’s comparison against a range that reaches 6K.1 Semgrep also names it as the only project able to run fully local and offline: an optional mode runs a security-tuned 8B-parameter model, Foundation-Sec-8B, entirely on-device, so no code leaves the machine. Semgrep frames that choice as depth and proof exchanged for breadth and speed, and recommends the tool to teams that need an offline or air-gapped scan over the broadest possible coverage.
Outputs / Products
- ai-deep-sast — fast SAST scanner combining Semgrep’s rule-based detection with LLM triage behind an evidence gate; a fully local fast-scan mode and a frontier-model deep-scan mode; Markdown, JSON, and JUnit output with CI severity gates.2
Positioning in the comparison
Semgrep’s capability matrix scopes ai-deep-sast to the OWASP/CWE Top 25, secrets, and AI/ML-specific findings across 30 or more languages, with an isolation posture of “n/a (static)” since the tool performs no dynamic execution.2 In the survey’s “finding” and execution tables, an ai-deep-sast finding is a triaged static match, and the tool produces no proof of concept or patch, offering advice only.
See Also
- ai-deep-sast — the product covered here.
- OSS AI Security Harness Comparison — source.