AWS's autonomous pentest agent exploits 33 of 37 fresh CVEs end-to-end
In short: AWS tested its AWS Security Agent (now part of AWS Continuum) against a freshly built benchmark of 37 real, recently disclosed CVEs, each run as a live containerized target with a hidden flag. Starting from a low-privilege endpoint with only a one-line hint and no proof-of-concept, the agent, running on Claude Opus 5, captured 33 of 37 flags through genuine end-to-end exploitation after an integrity audit, with 2 captures reclassified as unintended-path solves and 2 genuine misses where the agent broke the target before extracting the flag. AWS built this benchmark to avoid the contamination problem of public CVE suites, which models may have memorized from training data.
This summary was generated automatically by AI from AWS Security's publication. It is our own text, not a copy of the original — facts, figures and quotes belong to the source, linked above and below.
What changed?
- 1New contamination-reduced CTF-style benchmark of 37 recent CVEs (2025-2026) across 8 vulnerability classes, run with a 10-hour limit per sample
- 235 of 37 flags captured overall; 33 of 37 (89.2%) confirmed as genuine end-to-end exploitation via per-win integrity audit
- 32 captures reclassified as unintended-path solves after audit; challenges since hardened
- 42 genuine misses where the agent exploited the bug but crashed the target before reading the flag
- 5Validator layer confirmed 234 high-confidence findings, flagged 159 false positives, and deduplicated over 1,200 additional findings
- 6Median time to capture was approximately 4 hours; more than 6 captures involved multi-step chained exploits
Why it matters
This shows autonomous AI agents can now independently discover and weaponize real, unpatched-class vulnerabilities (RCE, SSRF, XSS-to-cookie-theft, SQLi, auth bypass) from minimal information, not just replay known exploits, which raises both offensive and defensive implications for any organization running web services.
What it means for AI agents and contact centers
If your company exposes web apps, APIs, admin consoles, or internet-facing services (including VoIP/SIP infrastructure or CRM integrations), this signals that automated AI-driven exploitation of freshly disclosed CVEs is becoming practical and fast (median ~4 hours), reinforcing the need for rapid patching, network segmentation, and monitoring of internet-facing endpoints rather than relying on obscurity or slow patch cycles.
Sources
- AWS SecurityOfficialPrimary sourceOriginal article →„AWS Security Agent (now part of AWS Continuum) exploited 33 of 37 new real-world CVEs“15 Oct 2026, 19:10
- Published by source
- 15 Oct 2026, 19:10
- Found by our system
- 15 Oct 2026, 19:11
- Summary generated
- 15 Oct 2026, 19:12
This article was written by AI from the original source. Facts, numbers and prices come from the source; missing values are marked “Not specified”. Legal notice, copyright and privacy