LIVE TRACKER — AI AS DISCOVERER Updated Nov 15, 2026 · Next check Mon ~1:50 PM PT · Weeks start Monday
Security flaws & hacks discovered by AI

7,613 flaws
AI found first.

This is the other side of the ARTEX tracker. There, AI carries out attacks. Here, AI is the discoverer — finding exploitable flaws in real software before attackers do. In 2026 the finds stopped arriving one at a time: one program disclosed 5,674 in a single week, and one model produced 948 working kernel exploits in 12 hours.

Counting rule: This tracker counts every single flaw instance that could be exploited by an attacker. A bulk find is not one event — GPT-6 Cyber's 948 kernel flaws count as 948. Only exploitable flaws count. Each distinct hack discovered by AI counts once — there were none in 2026.
Cumulative total · Exploitable flaws · 2026
7,613

Across 30 discovery events, Jan 1 – Nov 15, 2026. Dated by first public disclosure.

Conservative floor ≈ 7,482. Two large program aggregates are not itemized publicly, so up to ~32 individually-listed CVEs may already be inside them. See “Reading the total” below.

5,674
Week of Oct 5
Biggest week by far — Anthropic's OSS Scanner aggregate. 75% of the year's entire total in one disclosure.
948 in 12 hrs
GPT-6 Cyber · Linux
948 distinct kernel flaws, each with a working local privilege-escalation exploit. Disclosed Nov 4.
30 events
Discovery groups
16 of 45 weeks had any disclosure at all. The rest were zero — finds arrive in bursts, not a steady drip.
0 hacks
Hacks discovered by AI
No 2026 case found where AI discovered or exposed a hack/breach. AI found flaws — twice in software already being exploited.

Cumulative flaw total — 2026

This is a cumulative bar chart — a running total, not per-week counts. Each bar adds that week’s new flaws to everything before it, so the bars only rise or stay the same height and end at the full-year total of 7,613. Weeks are Monday-start, Jan 5 to Nov 15, 2026, including zero weeks. Switch to a Per week view to see individual weeks on their own. Hover / tap a bar for details.

Cumulative view — running total. The height of each bar is the total of all flaws discovered up to and including that week. Bars at the same height as the week before mean no new flaws that week; a taller bar means new flaws were added. The bars never get shorter, and the last one ends at 7,613.
Cumulative total (running sum) — one bar per week Zero week (no new flaws)Running total = 7,613 by week of Nov 9, 2026
↗
How to read these cumulative bars: The bars never get shorter — a bar the same height as the one before it means zero new flaws that week, not a fall. Each step up is that week’s new flaws added to the running total. The chart is also disclosure-dated, not discovery-dated. Aggregates like the OSS Scanner (found Nov 2025 – Oct 2026, announced Oct 8) and Big Sleep on V8 (found Apr–Sep, published Oct 27) land entirely in their announcement week. The shape shows when the world learned about the flaws — the underlying finding was spread over months.

By month

Same 7,613 flaws, grouped by disclosure month. October dominates because both big aggregates were announced then.

2026 · Jan
12
2026 · Feb
539
2026 · Mar
10
2026 · Apr
1
2026 · May
2
2026 · Jun
28
2026 · Jul
0
2026 · Aug
0
2026 · Sep
6
2026 · Oct
5,891
2026 · Nov*
1,124

* November is through Nov 15 — GPT-6 Cyber Linux (948, Nov 4), VicOne robotics (28, Nov 1) and N-able Able2Shell (30, Nov 8), plus Fortinet Mythos (18, Nov 10), Microsoft MDASH (33, Nov 11), Forescout (66, Nov 12) and Anthropic Apache CVE-2026-119618 (1, Nov 11), grouped by disclosure date.

Reading the total

Anthropic OSS Scanner aggregate5,674
OpenAI GPT-6 Cyber (Firefox 81 + Linux 948)1,029
Anthropic Opus 4.6 “500+” aggregate500
Everything else (27 events)410

Overlap warning

The 500+ aggregate may already include the Firefox CVEs (28) and FreeBSD CVE-2026-4747 (1). The 5,674 aggregate may already include Anthropic's two curl CVEs from Oct 14 (2) and the Mythos Linux CVE-2026-98432 (1). If all of those overlaps are real, the true total is ≈ 7,482, not 7,613. The Microsoft MDASH (33) and Forescout (66) finds may also already sit inside the 5,674 OSS Scanner aggregate — the kernel and embedded stacks overlap its scope — so the floor allows for those too. Anthropic has not published itemized lists, so it cannot be resolved from public sources. Individually-identified CVEs are counted separately here and flagged in place.

Unverified counts: two of the newest additions — VicOne's 28 robotics flaws and the N-able Able2Shell 30 — are counted as reported but have not been independently verified. Both counts may be revised in a later update.

What “exploitable” means here: every counted flaw was CVE-assigned, vendor-confirmed, externally validated, or shipped with a working exploit / reproducible PoC — depending on the entry. It does not mean a full weaponized exploit was published for each one. Entries where that standard is softer are flagged “Borderline” below with the reason. Unconfirmed candidates are never counted: the OSS Scanner's ~23,000 unreviewed candidates and Dinkin & Kraft's ~100 estimated driver flaws are both excluded.
The record

All 30 discoveries

Newest first · Number = exploitable flaws contributed

Dates are first public disclosure (CVE publication, vendor advisory, or researcher writeup). Campaign aggregates are dated at the disclosure of the aggregate and flagged — their flaws were found across the window stated in each entry.

Ruled out Nov 15

Not counted — and why

ZDI — 47 zero-days in Meta ML libraries · Nov 11, 2026

Discovered by human researcher Mat Powell (TrendAI ZDI) and reported to Meta in March 2026. Flaws in AI libraries, not found by AI. Out under the AI-is-the-discoverer rule.

AI-carried-out attacks — separate tracker

ARTEX and other agentic attack campaigns are attacks, not discoveries. They live on the ARTEX attack tracker and are never double-counted here.

Unconfirmed candidates & estimates

OSS Scanner ~23,000 unreviewed candidates; 4,824 direct-to-maintainer disclosures the dashboard itself warns may contain false positives; Dinkin & Kraft's ~100 estimated exploitable driver flaws (only the 15 confirmed are counted). Glasswing's “129,000 vulnerabilities found by AI” claim — no methodology, not counted.

Flaws in AI systems, found by humans

17 CVEs in AI agent frameworks (Oct 2026) are vulnerabilities in AI tools discovered by researchers — the direction is reversed, so they are out.

Evals, reproductions & AI-caused incidents

Benchmark scores with no real-world find, CVE-Genie reproducing known CVEs, sandbox escapes and other incidents the AI caused rather than discovered — all out. Real but uncounted finds (Mythos Safari / Android / Linux chains with no CVE counts published) are noted in the baseline log.

How this tracker counts

  • AI is the discoverer: the AI system did the finding — autonomously or with human direction. Human-led pentests where AI only marginally helped are out; genuinely AI-assisted finds are in and flagged.
  • Every flaw instance counts: 948 kernel flaws = 948. A discovery event is just the grouping that keeps the total auditable.
  • Exploitable only: theoretical or non-exploitable findings don't count. Confirmation standards vary by entry and are flagged where softer.
  • Hacks discovered count once each: a distinct hack/breach that AI discovered or exposed = 1. Total for 2026 so far: 0.
  • Weekly replacement: this card is replaced each Monday with the new total, the new week added to the chart, and any new discoveries appended above.