Chronicle/Capability & gating

Epoch AI reports GLM 5.3 exploited the patch gap in six Firefox releases in offline experiments

Oct 29, 2026CapabilityEvaluation reportSignificance assistant-drafted

Epoch AI reports that GLM 5.3 developed file-reading exploit chains for six Firefox releases using information public at each release date. Researchers supplied an offline setup with tools and materials, then let the model work without further human input; every chain also used an unpublished vulnerability that Mozilla patched in Firefox 159.0.

Why it matters

It links open-weight exploit capability to release engineering and the interval between public fixes and deployed updates.

Key facts

As stated in the sources, with where to find them.

Epoch reports about $1,000 marginal API cost per Firefox version after more than $100,000 total Firefox research spend. Equivalent Chrome attempts did not succeed at about $100,000 in API costs.Epoch AI, Results
The six releases span September 17–October 26. Epoch reports successful file reads and intermediate execution inside the web-content process, but no successful follow-on arbitrary-code-execution result. It had not confirmed equivalent attacks against Firefox 159.0 after the October 27 patch.Epoch AI, Results; notes 4–6
Mozilla describes weekly releases and withheld regression tests, and proposes coordinating fixes so patch gaps in separate security layers do not overlap. That structural change was planned, not claimed fully deployed.Mozilla, The patch gap; A new approach
Epoch says it paused all runs after monitoring flagged agents acting outside their sandbox in broken environments; those actions involved neither network access nor malicious activity.Epoch AI, Appendix: Safety measures for our experiments

Findings that cite this record

No tracked finding cites this record yet.

Key questions this bears on

Is AI shifting the balance between finding and fixing vulnerabilities?Discovery is ahead. Verification, triage, coordinated disclosure and safe repair remain bottlenecks between AI findings and deployed fixes.
How are attackers using AI agents in real operations?To run parts of intrusions at scale: vendors report a state-aligned actor and criminal groups using open-source agent frameworks and commercial models.

Sources

AI can turn security fixes into working exploits before users receive an updateEpoch AIOct 29, 2026Primary sourceAccessed Oct 30, 2026
Defending the patch gap against automated attackersMozillaOct 29, 2026Primary sourceAccessed Oct 30, 2026

Related records

Sep 29, 2026

Anthropic assesses open-weight GLM-5.3: exploit results near Mythos Preview, safeguards bypassed in 64% to 100% of simulated trials

CapabilityEvaluation reportAnthropic, US Center for AI Standards and Innovation, Moonshot AI

Oct 27, 2026

Anthropic describes five real-software exploit chains developed with an unreleased research model

CapabilityEvaluation reportAnthropic, Google, Apple

Jul 23, 2026

UK AISI and US CAISI jointly assess Kimi K3 cyber capability as trailing US frontier models

CapabilityEvaluation reportUK AI Security Institute, US Center for AI Standards and Innovation, Moonshot AI

Oct 28, 2026

Google announces Gemini 4 Carbon cyber results and phased access for defenders

CapabilityEvaluation reportGoogle DeepMind

Oct 12, 2026

Anthropic reports a 9B model sets up inference on 8 of 9 hardware types and GLM-5.2 completes simplified replication

CapabilityEvaluation reportAnthropic

Oct 13, 2026

CyberArena benchmark pits 12 frontier agents against each other in live attack-and-defense matches over 466 games

CapabilityBenchmarkUC Berkeley, UC Santa Cruz, Purdue University

Organizations: Epoch AI, Zhipu AI (Z.ai), Mozilla
Topics: Exploit development Open-weight diffusion Capability evaluation Vulnerability discovery
Attack surface: Agent model Execution sandbox Evaluation environment
Methods: AI-assisted vulnerability exploitation AI vulnerability discovery for defense
Benchmarks and tools: GLM (Zhipu AI)
Fide research: FID-075: Cybersecurity capabilities and whole-system risk
Record: epoch-glm-firefox-patch-gap-2026 · added 2026-10-30

Cite this record

Fide AI. "Epoch AI reports GLM 5.3 exploited the patch gap in six Firefox releases in offline experiments." Agentic Cyber Explorer, record epoch-glm-firefox-patch-gap-2026. https://agentic-cyber-explorer.pages.dev/events/epoch-glm-firefox-patch-gap-2026/ (corpus as of 2026-10-30).

BibTeX

@misc{fide-ace-epoch-glm-firefox-patch-gap-2026,
  author = {{Fide AI}},
  title = {Epoch AI reports GLM 5.3 exploited the patch gap in six Firefox releases in offline experiments},
  howpublished = {Agentic Cyber Explorer},
  year = {2026},
  url = {https://agentic-cyber-explorer.pages.dev/events/epoch-glm-firefox-patch-gap-2026/},
  note = {Record epoch-glm-firefox-patch-gap-2026; corpus as of 2026-10-30}
}

Data: JSON · CC BY 4.0

Spotted an error? Suggest a correction.