Threat reportVulnerabilityTL-2026-3489

Full-chain Chrome exploit (V8 Maglev, WebP CVE-2026-85045, Crashpad CVE-2026-78952) built with GPT-5.6 Sol by XBOW earns $250,000 Google bounty

highPATCHED

Full-chain Chrome exploit (V8 Maglev, WebP CVE-2026-85045 (TL-2026-3489), also tracked as Zero to Calc in Six Days, is a high-severity software vulnerability scored CVSS 8.3, first published 2026-10-14. It has no confirmed attribution, affects Google Chrome (Windows), references 2 CVEs (CVE-2026-85045, CVE-2026-78952), maps to 10 MITRE ATT&CK techniques (T1059.003, T1059.007, T1068), and is covered by 9 detection rules and 20 indicators of compromise.

CVSS
8.3/10High
CVEs
2Referenced vulnerabilities
Techniques
10MITRE ATT&CK
Actors
0Not attributed
Detection rules
9SPL · KQL · Sigma
IOCs
20Indicators of compromise

Key facts for TL-2026-3489

Threat ID
TL-2026-3489
Also known as
Zero to Calc in Six Days, XBOW Chrome Full Chain
Severity
HIGH
CVSS
8.3 (CVSS:3.1/AV:N/AC:H/PR:N/UI:R/S:C/C:H/I:H/A:H)
Status
PATCHED
Category
VULNERABILITY
First published
Last reviewed
Attribution confidence
LOW
Motivation
UNKNOWN
Target sectors
technology, government administration, finance, health, education
Target regions
Global
Detection rules
9
Indicators of compromise
20
Updates
2026-10-14 · 2 updates · revalidated 2× · latest source

Malware and tooling in Full-chain Chrome exploit (V8 Maglev, WebP CVE-2026-85045

Malware and tooling: gpt-5.6-sol

How Full-chain Chrome exploit (V8 Maglev, WebP CVE-2026-85045 works

XBOW researchers, working with OpenAI's gpt-5.6-sol model, built a Windows Chrome full-chain exploit in six days (about $40,000 of API compute) by chaining a V8 Maglev deoptimization bug, an animated-WebP decoder out-of-bounds write that escapes the V8 sandbox, and a Crashpad bug that escapes the renderer sandbox. Google fixed all three flaws and paid a $250,000 bounty; no in-the-wild exploitation has been reported.

XBOW published its write-up, 'Zero to Calc in Six Days: Inside XBOW's Chrome Full Chain', on 2026-10-13. A human researcher directed OpenAI's publicly available gpt-5.6-sol model between 2026-08-02 and 2026-08-07 and spent roughly $40,000 in API costs. The result was a three-stage chain from a malicious web page to code execution outside Chrome's renderer sandbox on Windows. Google's Chrome VRP paid $250,000. Press coverage describes this as the first full-chain award under the April 2026 VRP rules and the first granted to an autonomous platform. XBOW estimates the affected population at about one billion Chrome-on-Windows users. All three bugs were reported through Google's official channels and are fixed. No exploitation of real users has been reported.

Stage 1, V8 Maglev (Chromium issue 547819997, V8 change 8255554; no CVE identified in the sources reviewed). Maglev reused a mutable boxed number (HeapNumber) during deoptimization, which broke the invariant that HeapNumber deduplication applies only to immutable HeapNumbers. The deoptimizer materialized one mutable object that was then shared across separate JavaScript values. The exploit used this to forge metadata for a Uint8Array and gain read/write access across the 4GB pointer-compression cage. This is the V8 heap cage, not the V8 sandbox.

Stage 2, animated WebP decoder (Chromium issue 548509153; Chromium changes 8266257 and 8352899). Press reporting maps this stage to CVE-2026-85045, which NCSC-NL's CSAF record describes as a V8 race condition (CWE-367, CVSS 7.5) fixed in Chrome 152.0.7977.82. The XBOW write-up does not state CVE IDs, so the mapping is taken from the XBOW press release. This is a time-of-check/time-of-use flaw. The decoder cached frame geometry from one image while reading bytes from a substituted image (via a transferred ArrayBuffer), producing a heap overflow. Modified WebP geometry and payload gave a single attacker-chosen 64-bit native write, which escapes the V8 sandbox. The chain then used a Wasm Code Pointer table base modification and an 'island-hopping' loader of 13 64-bit immediates in Liftoff-generated Wasm code, joined by short jumps (about 70 bytes), to redirect execution. The write-up says CFG and CET were partially circumvented through valid call sequences to existing functions.

Stage 3, Crashpad on Windows (Chromium issue 545820931; Chromium change 8267322; CVE-2026-78952, CWE-787, CVSS 8.3, fixed in Chrome 152.0.7977.65). Crashpad's retry logic after a failed memory read assumed the target was unchanged between the failed and retried read. A freeze-exempt renderer thread could flip page protections during snapshot creation and cause an oversized read and a 0xa8-byte buffer overflow. That overflow corrupted the vtable pointers and state fields of a live SnapshotMinidumpMemoryWriter. Redirecting the memory-read callback to FallbackUnhandledExceptionFilter ran `cmd /k calc.exe` outside the sandbox, which demonstrates arbitrary program launch.

Defenders should note the real risk: AI assistance sharply lowers the cost and time to build a browser full chain, and XBOW says defenders should plan accordingly. The sources provide no network or file IOCs and no in-the-wild activity. The IOC list below holds only behavioral and artifact indicators taken from the write-up. CVE-2026-85046 (V8 type confusion, exploited in the wild, Chrome 152.0.7977.82) is a separate bug reported by another researcher and is NOT part of this chain. The cvss_score and vector shown are those of the highest-rated component CVE (CVE-2026-78952), not a rating of the chain. XBOW's V8 bug has no CVE in the reviewed sources. The mitigations are in Chrome 152.0.7977.65 and later: V8 now blocks HeapNumber deduplication during deoptimization, the WebP decoder validates stride/size against actual bitmap dimensions, and Crashpad clamps the retry length to the minimum of the page remainder and the original request size. Exploit code release was not confirmed in the sources.

MITRE ATT&CK techniques used in TL-2026-3489

Execution

T1059.003 Windows Command Shell; T1059.007 JavaScript; T1106 Native API; T1203 Exploitation for Client Execution

Privilege Escalation

T1068 Exploitation for Privilege Escalation

Initial Access

T1189 Drive-by Compromise

Defense Evasion

T1211 Exploitation for Defense Evasion; T1620 Reflective Code Loading

Resource Development

T1587.004 Exploits; T1588.007 Artificial Intelligence

Affected products and versions in Full-chain Chrome exploit (V8 Maglev, WebP CVE-2026-85045

  • Google — Chrome (Windows)
    Vulnerable versions: prior to 152.0.7977.65 (Crashpad CVE-2026-78952); prior to 152.0.7977.82 (V8/WebP CVE-2026-85045)
    Fixed in: 152.0.7977.82 and later
  • Microsoft — Edge (Chromium)
    Vulnerable versions: prior to 152.0.4191.65 (CVE-2026-85045)
    Fixed in: 152.0.4191.65 and later
  • Debian — Chromium
    Vulnerable versions: prior to 152.0.7977.82-1~deb12u1 (CVE-2026-85045)
    Fixed in: 152.0.7977.82-1~deb12u1

Remediation for Full-chain Chrome exploit (V8 Maglev, WebP CVE-2026-85045

Patches

  • Chrome 152.0.7977.65 (CVE-2026-78952, Crashpad)
  • Chrome 152.0.7977.82 (CVE-2026-85045 and later fixes)
  • Microsoft Edge 152.0.4191.65
  • Debian Chromium 152.0.7977.82-1~deb12u1

Immediate actions

  • Update Google Chrome on Windows to 152.0.7977.82 or later (Crashpad fix landed in 152.0.7977.65) and relaunch the browser so the update takes effect
  • Update Microsoft Edge to 152.0.4191.65 or later and other Chromium-based browsers to their equivalent fixed builds
  • Verify browser version compliance across the Windows fleet with endpoint management tooling

Workarounds

  • No vendor workaround documented; apply the patched browser builds

Longer-term hardening

  • Treat AI-assisted exploit development as lowering the cost of browser full chains and shorten browser patch SLAs accordingly
  • Monitor crashpad_handler.exe for child process creation or unexpected module loads
  • Enable enterprise browser auto-update and enforce relaunch deadlines

CVEs associated with Full-chain Chrome exploit (V8 Maglev, WebP CVE-2026-85045

CVE-2026-85045, CVE-2026-78952

Weaknesses (CWE) in Full-chain Chrome exploit (V8 Maglev, WebP CVE-2026-85045

CWE-367, CWE-787

Timeline of Full-chain Chrome exploit (V8 Maglev, WebP CVE-2026-85045

  • Challenge to build a Chrome full chain set at an XBOW Malta staff meeting (January 2026; exact day not stated, date is a month placeholder)
  • Google's new Chrome VRP rules take effect (April 2026; exact day not stated, date is a month placeholder); XBOW's payout is the first full-chain payout under them.
  • XBOW completes a first full Chrome exploit chain after roughly two months of work (June 2026; exact day not stated, date is a month placeholder); the Google-awarded chain was a later, separate build
  • XBOW researcher begins building the Chrome full-chain exploit with gpt-5.6-sol (Sunday)
  • XBOW's initial Gin-based V8 sandbox escape stage collided with a duplicate report (Chromium issue 542670566) and was replaced within hours by the animated-WebP stage; Google confirmed the chain.
  • Six-day development window ends (Friday); chain demonstrated launching calc.exe outside the renderer sandbox on Windows, at about $40,000 in API cost
  • CVE-2026-78952 (Crashpad out-of-bounds write, CVSS 8.3) published; fixed in Chrome 152.0.7977.65
  • CVE-2026-85045 (V8 race condition / TOCTOU, CVSS 7.5) initially released; fixed in Chrome 152.0.7977.82 and Edge 152.0.4191.65
  • XBOW publishes 'Zero to Calc in Six Days' and announces the $250,000 Chrome VRP reward, the first full-chain award under the 2026 rules
  • Cyber Security News and other outlets report that all three flaws are fixed and no real-world exploitation is confirmed

Update history for TL-2026-3489

Sources cited for Full-chain Chrome exploit (V8 Maglev, WebP CVE-2026-85045

Detection coverage for TL-2026-3489

As of 2026-10-14, Threadlinqs Intelligence publishes 9 detection rule(s) for TL-2026-3489 across Splunk SPL, Microsoft KQL and Sigma, covering 20 indicator(s) of compromise. The whole corpus is readable without an account; a free account unlocks full detection query text in Splunk SPL, Microsoft KQL and Sigma; paid tiers add raw indicator values, correlation and the MCP server. Threadlinqs MCP server · View plans.

9 detection rules (Splunk SPL, Microsoft KQL, Sigma) · Blue and above. Compare plans
20 indicators of compromise · Red and above. Compare plans

Further reading

Threadlinqs Intelligence — Real-Time Threat Detection Platform

[ 0 threats ] [ 0 det ] [ CRIT: 0 ] [ HIGH: 0 ]
// threat_feed
$ sort --newest
Showing all threats

Live intelligence console

Threat level
Fig. 01 · Threat weatherIndexing the archive…
1 square = 1 threat · click to open

Latest Threats