Skip to content

Topic

AI-Assisted Vulnerability Discovery

The use of large language models and autonomous agents to automatically find software vulnerabilities, rivaling manual security research.

Current stories

leadershipConfirmed23 publishers

Microsoft's October release of 1,407 CVEs tops 2020's full-year record by 146

Microsoft shipped a record 1,407 CVEs on October Patch Tuesday, including an exploited Windows Shell Mark of the Web bypass. The component's 2026 record argues for a standing slot in the monthly patch cycle.

Perspective Coverage

23 publishers
Builder
Builder 16%
Operator
Operator 68%
Investor
Investor 16%

Reality

Evidence74
Adoption
Not scored
Hype gap+28
Incentives
Not scored
Confidence72
securityOne report1 publisher

OpenSSH moves to more frequent releases after AI-found bugs turn up again from other researchers

OpenSSH will release more often because bugs found by AI tools are being rediscovered by other researchers, its maintainers said with version 10.6. The team takes that as a sign that attackers who never report bugs can find the same flaws.

Publishers:seclists.org

Reality

Evidence65
Adoption
Not scored
Hype gap+5
Incentives
Not scored
Confidence62
securityOne report1 publisher

Thales's own test shows its new binary protection cutting an AI agent's bug finds from eight to zero

Thales says its new Sentinel Envelope Plus binary protection cut an AI agent's finds in one test application from 8 of 10 vulnerabilities to zero. The bugs stay in the code, so software vendors would be buying time to patch, and the only measurement of that time so far comes from the seller.

Reality

Evidence25
Adoption
Not scored
Hype gap+45
Incentives85
Confidence35
securityConfirmed4 publishers

Google traces most of 2026's exploitation growth to fast n-day weaponization

Google's threat intelligence group counts 18 exploited flaws a month in 2026, up from 10.5 in 2025, while zero-days rose only from eight to 11. GTIG attributes most of the added attacks to fast weaponization of disclosed n-days, so the exposure sits in the days after a patch ships.

Perspective Coverage

4 publishers
Builder
Builder 33%
Operator
Operator 61%
Investor
Investor 6%

Reality

Evidence72
Adoption
Not scored
Hype gap+30
Incentives35
Confidence65
productOne report1 publisher

Oracle pushes an agent's read limits down to the row, column and cell

Oracle's database chief says AI is turning up hundreds of security issues in code his teams spent decades hardening. His prescription for agents is authorization enforced inside the database, keyed to the end user's identity.

Publishers:siliconangle.com

Reality

Evidence38
Adoption12
Hype gap+30
Incentives88
Confidence45
productOne report1 publisher

Microsoft's record 974 patches are one month inside a year that has logged 66,401 CVEs

The year's CVE tally has reached 66,401, close to double where it stood last September, and Microsoft alone patched 974 in a single month. Jerry Gamblin, who keeps the count, says more known bugs is mostly the system working.

Publishers:wired.com

Reality

Evidence55
Adoption45
Hype gap+25
Incentives45
Confidence50
securityOne report1 publisher

Microsoft expects 58 of September's 973 CVEs to be exploited within 30 days

Microsoft rated 114 of the 973 Critical, but 284 score 8.0 or higher and two Important-severity Windows bugs are already under attack. One 9.1 advisory in the same cycle came from outside Microsoft.

Reality

Evidence62
Adoption55
Hype gap+12
Incentives68
Confidence58
productOne report1 publisher

An OpenAI model found an 8.8 token-replay flaw in code for an EU project

Politico reported that ENISA and CERT-EU ran an advanced OpenAI model over an EU project's code and got four fixed flaws out of it. Poland's CERT, doing the same kind of work, said it tested every hypothesis on real systems.

Publishers:thenextweb.com

Reality

Evidence58
Adoption62
Hype gap+16
Incentives71
Confidence52
productOne report1 publisher

Confirmed exploitation now lands 40 days sooner after a CVE goes public

VulnCheck's first-half figures put the median at 80 days, down from 120 in 2025, while the count of CVEs exploited within a month of publication held flat near 200. The tier that breaks is the 90-day one.

Publishers:vulncheck.com

Reality

Evidence58
Adoption62
Hype gap−12
Incentives70
Confidence55
investConfirmed3 publishers

Ten days of maintainer time bought Core Lightning a two-week embargo on AI-found bugs

Core Lightning says the machine-written reports found real flaws in its payments software. It is holding the count and severity for at least two weeks, leaving node operators to decide what to do with a hazard whose size is still private.

Perspective Coverage

3 publishers
Builder
Builder 28%
Operator
Operator 52%
Investor
Investor 20%

Reality

Evidence60
Adoption40
Hype gap+25
Incentives68
Confidence62
productOne report1 publisher

An Apple fleet admin would lock unpatched Macs by end of day to keep pace with AI-found bugs

Bradley Chambers argues in his sponsored 9to5mac column that AI-assisted vulnerability discovery has killed the 90-day compatibility test, and that Mac fleets now need device management willing to stop work to force a patch.

Publishers:9to5mac.com

Reality

Evidence30
Adoption20
Hype gap+42
Incentives75
Confidence58
productConfirmed3 publishers

Extended Stable now bundles four Chrome milestones into the same eight-week window

Chrome 153 started a two-week milestone cadence, which leaves the enterprise Extended Stable channel on its old eight-week calendar carrying twice as much change per jump. Google wants most admins on Stable anyway.

Perspective Coverage

4 publishers
Builder
Builder 39%
Operator
Operator 45%
Investor
Investor 16%

Reality

Evidence70
Adoption68
Hype gap+18
Incentives74
Confidence76
leadershipOne report1 publisher

Cheap AI bug hunting collides with the one step that still needs engineers

A J.P. Morgan report counts more than 1,400 AI-generated vulnerability reports that maintainers accepted at roughly 90% validation and mostly did not patch, which moves the budget question from finding to fixing.

Publishers:forbes.com

Reality

Evidence28
Adoption32
Hype gap+42
Incentives85
Confidence60
productOne report1 publisher

Nine patched flaws in CryptoPro Secure Disk turn pre-boot encryption into an inventory problem

CryptWare shipped fixes in two releases without publishing update notes, and the ATM maker that embeds the software says only two bugs touched its product. Who else runs it is the harder question, and the answer is not written down anywhere.

Publishers:wired.com

Reality

Evidence64
Adoption31
Hype gap+12
Incentives66
Confidence55
investOne report1 publisher

Coinkite now makes Coldcard owners roll dice, after $130M walked out of air-gapped wallets

A 2021 seed flaw cut entropy from 128 bits to about 40, so no attacker needed to touch the hardware. Every seed generated before July 2026 now has to be replaced.

Publishers:decrypt.co

Reality

Evidence62
Adoption54
Hype gap+14
Incentives66
Confidence58
scienceOne report1 publisher

Microsoft's AI bug finders have outrun its Exchange engineers, and CU1 is the casualty

Microsoft has delayed the first cumulative update for Exchange Server Subscription Edition a second time, with no new date, while engineers validate a growing pile of AI-found security findings.

Reality

Evidence58
Adoption24
Hype gap+22
Incentives62
Confidence55
securityOne report1 publisher

The bottleneck moved: 622 CVEs in July, and no one left to write up the fixes

A Black Hat USA 2026 keynote described AI tooling that found roughly 1,000 bugs and then stalled on reporting. Patch Tuesday volume tells the same story from the other end.

Reality

Evidence30
Adoption24
Hype gap+34
Incentives54
Confidence30