AISLE — 12 of 12 OpenSSL Vulnerabilities

AISLE Discovered 12 out of 12 OpenSSL Vulnerabilities

Source: AISLE blog — AISLE Discovered 12 out of 12 OpenSSL Vulnerabilities (January 2026). Author: Stanislav Fort. Contributing researchers: Petr Šimeček, Tomas Dulka, Luigino Camastra. Local copy: .raw/articles/aisle-12-of-12-openssl-vulnerabilities-2026-01.md.

Key Claim

AISLE’s autonomous analyzer discovered all 12 vulnerabilities in the January 2026 coordinated release of OpenSSL — including CVE-2025-15467, a CVSS 9.8 stack-buffer-overflow in CMS AuthEnvelopedData parsing whose vulnerable code dates to 1998. Span: 8+ subsystems (CMS, QUIC, TLS 1.3, post-quantum ML-DSA signatures, PKCS#12, OCB, TimeStamp, PKCS#7). 5 of 12 fixes were authored by AISLE and incorporated directly into OpenSSL. 6 additional findings were caught before the vulnerable code ever appeared in a release — the “preventing vulnerabilities, not merely patching them after deployment” framing.

The 12 CVEs

High and Moderate Severity

CVE	Severity	Subsystem	Description
CVE-2025-15467	High (NIST CVSS v3 9.8)	CMS	Stack buffer overflow in CMS AuthEnvelopedData parsing — potential remote code execution under specific conditions. Vulnerable code dates to 1998.
CVE-2025-11187	Moderate	PKCS#12	PBMAC1 parameter-validation missing → stack-based buffer overflow

Low Severity (10 CVEs)

CVE	Subsystem	Description
CVE-2025-15468	QUIC	Crash in QUIC protocol cipher handling
CVE-2025-15469	Post-quantum signatures (ML-DSA)	Silent truncation bug
CVE-2025-66199	TLS 1.3	Memory exhaustion via certificate compression
CVE-2025-68160	I/O	Memory corruption in line-buffering — affects code back to OpenSSL 1.0.2
CVE-2025-69418	OCB mode	Encryption flaw on hardware-accelerated paths
CVE-2025-69419	PKCS#12	Memory corruption in character encoding
CVE-2025-69420	TimeStamp	Crash in TimeStamp Response verification
CVE-2025-69421	PKCS#12	Crash in PKCS#12 decryption
CVE-2026-22795	PKCS#12	Crash in PKCS#12 parsing
CVE-2026-22796	PKCS#7	Crash in PKCS#7 signature verification — affects code back to OpenSSL 1.0.2

Three of the twelve bugs date to 1998–2000. They survived millions of CPU-hours of fuzzing by organizations including Google.

Notable Findings

• Watershed moment for AI-driven security on cryptographic libraries. OpenSSL is “one of the most deployed, battle-tested, and carefully maintained open-source projects in existence”; even a single accepted vulnerability per release represents a rare achievement. 12 in one coordinated release from a single research team — let alone an AI-driven one — is structurally unusual. Per Bruce Schneier on AISLE’s site: “AISLE is credited for surfacing 13 of 14 OpenSSL CVEs assigned in 2025, and 15 total across both releases. This is a historically unusual concentration.”
• AISLE proposed fixes adopted directly for 5 of 12 CVEs. The blog frames this as collaboration rather than analyst-throw-over-fence: AISLE submits reproduction steps + root-cause analysis + concrete patch proposals, and in each case its proposed fixes “either informed or were directly adopted by the OpenSSL team.”
• 6 additional findings caught pre-release — never assigned a CVE because the fixes were merged before the vulnerable code appeared in a release. This is the strongest single signal in the disclosure that the AISLE workflow is operationalized continuously inside the development workflow, not just as point-in-time audit.
• Time-to-remediation collapse. Stanislav Fort: “When autonomous discovery is paired with responsible disclosure, it collapses the time-to-remediation for the entire ecosystem.” This is the same TTR-collapse argument the Zero Day Clock frames empirically — but on the defender side rather than the attacker side. Defender-side TTR has historically been the bottleneck; AISLE’s claim is that AI-driven defender TTR can compress as fast as the attacker-side TTE has.
• Pairings with OpenSSL Foundation leadership. Tomáš Mráz (CTO, OpenSSL Foundation): “This release is fixing 12 security issues, all disclosed to us by AISLE. We appreciate the high quality of the reports and their constructive collaboration with us throughout the remediation.” Matt Caswell (Executive Director, OpenSSL Foundation): “Keeping widely deployed cryptography secure requires tight coordination between maintainers and researchers. We appreciate AISLE’s responsible disclosures and the quality of their engagement across these issues.” Maintainer-side validation is the structural difference between AI-driven findings and AI-driven value.

Strengths and Weaknesses

Strengths. Cryptographic libraries are the hardest domain to claim novel findings in — “finding a genuine security flaw in OpenSSL is extraordinarily difficult.” The 12-of-12 + 1998-vintage + 8-subsystem-span result establishes that AI-driven analysis can produce findings in a domain where traditional SAST has historically been brittle (complex logic errors, timing-dependent issues, deep subsystem interactions). The collaborative model (reproduction steps + root-cause analysis + patch proposals) is the production-grade version of AI-driven disclosure: the analyst gets actionable evidence and a starting fix, not just a finding to triage. The 6 pre-release catches are the strongest signal of operational integration.

Weaknesses and open scope.

• The autonomous analyzer architecture is not disclosed. Unlike OpenAnt (six-stage pipeline), Codex Security’s Aardvark (four-stage pipeline), or MDASH (five-stage with named agent classes), AISLE has not published its harness internals. The wiki’s cross-product FP-control mechanism comparison cannot yet place AISLE — see the org page for the mechanism-category gap.
• No false-positive rate published. AISLE reports 12 CVEs assigned and 6 pre-release catches; no comparable accept/reject ratio or per-finding triage cost is captured in this blog.
• No third-party benchmark. Comparisons to MDASH (88.45% on CyberGym), raw Mythos (83.1% on CyberGym), or Aardvark (92% on internal golden repos) require a common evaluation surface; the wiki’s longstanding gap on this remains.
• OpenSSL is the focused target. AISLE’s other surfaced disclosures (curl, FreeBSD) suggest broader OSS-cryptographic-and-systems scope, but the systemic coverage map across other widely-deployed OSS targets is not yet published.

Relations

• Supports Frontier AI for Vulnerability Discovery thesis — AISLE is now the wiki’s sourced anchor for AI-driven discovery of decade-class latent vulnerabilities in widely deployed open-source cryptographic libraries. Pairs with Glasswing’s 27-year-old OpenBSD vuln + 16-year-old FFmpeg vuln disclosures.
• Strengthens Mythos-ready briefing — the AISLE-OpenSSL data point is cited inline by the briefing’s timeline; the dedicated AISLE org page and this paper page now close that gap.
• Cross-references [[unprompted-conference-march-2026|[un]prompted Conference March 2026]] — AISLE is named in the conference’s overall key-claim alongside FENRIR, Promp2Pwn, and XBOW as systems running autonomous bug-finding agents at production scale.
• Adjacent to CyberGym Benchmark — UC Berkeley team’s Open-Ended Discovery mode published 35 zero-days + 10 unique zero-days at 969-day mean persistence; AISLE’s 12-of-12 OpenSSL result sits in a similar conceptual frame but on a specific high-value target rather than across a benchmark corpus.

AISLE's 6 pre-release catches are the operational signal

The 12 assigned CVEs are the publicly visible artifact. The 6 additional findings caught before the vulnerable code ever appeared in a release are the structural innovation — they document operational integration of autonomous analysis into the development workflow. This is the AI-driven realization of what the Mythos-ready playbook calls PA 1: Point Agents at Your Code and Pipelines, with the additional commitment that “all code (human or AI-generated) should pass LLM-driven security review before merge” — applied at the OpenSSL upstream-development level rather than at the consumer-application level. Pre-release catches are the no-CVE-needed outcome the discipline is designed to produce.