Thomas Mosk’s observational preprint estimates more than doubled public vulnerability disclosures for Glasswing-access firms relative to comparison firms. Measured remediation did not slow among fixed-and-disclosed flaws. A suggestive decline in the exploited share and no statistically detectable count change do not establish fewer attacks; non-random access and limited disclosure visibility constrain the conclusions.
Why it matters
It adds an independent attempt to measure security outcomes while exposing what public disclosure data cannot establish about unseen backlogs or attacks.
Key facts
As stated in the sources, with where to find them.
- The study compares selected Glasswing-access firms with firms without access, including Chinese and Russian firms, using public vulnerability disclosures. Access was not randomly assigned.Abstract; empirical design
- The author estimates more than twice as many disclosures and about 150% more high-impact vulnerabilities for access firms relative to comparison firms. These are observational estimates, not a count of flaws independently proven to have been found by AI.Abstract; results
- Measured vendor and distribution remediation did not slow among fixed-and-disclosed flaws. This selected sample cannot exclude a growing undisclosed backlog or establish when users installed fixes.Section 4: Remediation: was there a human bottleneck?
- The author estimates a roughly two-thirds decline in the share exploited within one month against the Chinese comparison, but calls the decline suggestive. Exploited-vulnerability counts show no statistically detectable change; imprecise estimates allow moderate changes in either direction. This does not establish fewer attacks.Section 6: exploitation analysis and inference caveats
- Program selection, national disclosure practices and incomplete vulnerability visibility limit interpretation. Comparisons are between Mythos access and the AI tools available to comparison firms, not AI versus no AI; additional comparison groups and robustness checks do not make access random.Introduction limitations; identification and robustness
Findings that cite this record
Key questions this bears on
- Is AI shifting the balance between finding and fixing vulnerabilities?Discovery is ahead. Verification, triage, coordinated disclosure, safe repair and getting fixes to users remain bottlenecks between AI findings and protection.
Sources
Related records
May 22, 2026
Nov 4, 2026
Oct 28, 2026
Oct 12, 2026
Oct 6, 2026
Sep 30, 2026