Vulnerabilities & CVEs

Chrome 147 Patches 60 Vulns, 2 Critical in WebML

Chrome 147 dropped with patches for 60 vulnerabilities, but the real shocker? Two critical WebML holes that netted researchers $86,000. This isn't routine housekeeping—it's a warning about browser ML's fragile underbelly.

Chrome 147 update screen showing security patches for WebML critical vulnerabilities

Key Takeaways

  • Chrome 147 patches 60 vulns, two critical WebML bugs earning $86K bounties for potential sandbox escapes.
  • WebML's push for on-device AI creates new attack surfaces, echoing past WebGL vulnerabilities.
  • Update immediately; new cookie protections add account security amid the fixes.

Everyone figured Chrome’s monthly updates would chug along like always—patch a dozen highs, sprinkle in some mediums, keep the web humming. But Chrome 147? It slams the door on 60 vulnerabilities, two of them critical enough to bag researchers $86,000 in bounties. That’s not just a payday; it’s a flare signaling deeper cracks in how browsers handle machine learning right in your tab.

Chrome 147. There it is, the keyword that should have every dev and security wonk hitting update. Google’s first stable drop this cycle lands with WebML—the browser’s on-device ML engine—taking the hardest hits.

What Makes These WebML Bugs So Nasty?

Heap buffer overflow in CVE-2026-5858. Integer overflow in CVE-2026-5859. Anonymous hunters spotted them, cashed $43K each. Sounds technical? Picture this: you’re running an ML model for image recognition or voice processing, all client-side to dodge server lag. But overflow those buffers, and attackers rewrite memory. Boom—sandbox escape, remote code execution. Your tab’s now a launchpad for malware.

Google rates them critical for a reason. WebML’s architected for speed—WebGPU backends, TensorFlow.js integrations—but speed skips checks. It’s like flooring it on a racetrack with no guardrails.

The critical vulnerabilities both impact Chrome’s WebML component, which is designed for running machine learning models directly in the browser.

That’s straight from Google’s advisory. Chilling, right? Because WebML isn’t fringe; it’s the future Google pushes hard—edge AI without phoning home.

And here’s my take, one you won’t find in the press release: this echoes WebGL’s rocky 2010s start. Back then, 3D graphics in browsers birthed a vulnerability goldmine—Skia flaws, ANGLE mishaps. WebML? Same playbook. As browsers gobble more compute-heavy features, attack surfaces balloon. Prediction: by 2026, we’ll see WebML zero-days weaponized in the wild, especially with AI hype drawing script kiddies.

Short para punch: Update. Now.

The rest? 14 highs across V8 (JS engine), WebRTC (video calls), Blink (rendering core). Google’s internals snagged nearly half—impressive, but anonymous reports fill the gaps. Bug bounties trickle out: $11K for a PrivateAI use-after-free (CVE-2026-5874), smaller pots elsewhere.

No in-the-wild exploits announced. Yet. Remember late March’s zero-day frenzy? Chrome patched 21, one actively exploited. Pattern’s clear—patch fast, or pay later.

Why Does WebML Keep Bleeding Criticals?

Look, WebML’s ambitious. It ports ONNX models, taps WebGPU for parallelism. But integer overflows? That’s math gone wrong in tensor ops. Heap overflows? Bounds-checking fails during inference. Why? Rushed implementations chasing TensorFlow parity. Google’s sprinting to own browser AI, but corners cut.

Corporate spin calls it ‘proactive security.’ Please. $86K screams ‘we almost got owned.’ Skepticism’s warranted—WebML’s opt-in, sure, but bundled in Chrome. One bad site, and you’re pwned.

Bonus in 147: session cookie shields. Stolen auth cookies? Chrome now rotates ‘em harder, thwarts takeovers. Smart, under-the-radar win amid the vuln storm.

Should You Rush to Chrome 147 Today?

Yes. Auto-updates lag on some setups—enterprise locks, slow channels. Manual check: chrome://settings/help. Restart. Done.

But dig deeper. This update exposes architecture shifts. Browsers aren’t document viewers anymore; they’re OSes-in-waiting. V8 for compute, WebAssembly for ports, now WebML for models. Each layer adds vectors. Google’s Blink monopoly (70% share) means one flaw ripples global.

Wander a sec: internals fixed 30-ish. External eyes caught the gems. Bug bounty’s working, but scale it—pay more for WebGPU next.

Medium para: Highs in WebAudio, Media, ANGLE. Familiar foes. Skia’s vector graphics still bites.

One medium stands out—no bounty fanfare, but a PrivateAI use-after-free? AI privacy tools ironically holed.

The Bigger Picture: Browser AI’s Security Debt

Google’s not alone. Safari, Firefox chase ML parity. But Chrome leads, so it bleeds first. Historical parallel: Flash’s deathbed vulns as HTML5 rose. WebML could force a reckoning—sandbox it tighter, or fork to extensions?

Bold call: expect regulatory heat. EU’s DMA eyes gatekept APIs; vulns like these fuel ‘open up Chrome’ cries.

Wrapping the dive—Chrome 147’s a bandage on growing pains. Update, watch WebML, question the rush to browser brains.


🧬 Related Insights

Frequently Asked Questions

What are the critical vulnerabilities in Chrome 147? Two WebML flaws: heap buffer overflow (CVE-2026-5858) and integer overflow (CVE-2026-5859), each worth $43K bounties. They risk sandbox escapes.

Do I need to update to Chrome 147 immediately? Absolutely—60 patches, no known exploits yet, but criticals demand it. Check chrome://settings/help.

What is WebML and why is it vulnerable? Chrome’s tool for browser ML models. High-speed design skips safety, leading to overflows in tensor processing.

James Kowalski
Written by

Investigative tech reporter focused on AI ethics, regulation, and societal impact.

Frequently asked questions

What are the critical vulnerabilities in Chrome 147?
Two WebML flaws: heap buffer overflow (CVE-2026-5858) and integer overflow (CVE-2026-5859), each worth $43K bounties. They risk sandbox escapes.
Do I need to update to Chrome 147 immediately?
Absolutely—60 patches, no known exploits yet, but criticals demand it. Check chrome://settings/help.
What is WebML and why is it vulnerable?
Chrome's tool for browser ML models. High-speed design skips safety, leading to overflows in tensor processing.

Worth sharing?

Get the best Cybersecurity stories of the week in your inbox — no noise, no spam.

Originally reported by SecurityWeek

Stay in the loop

The week's most important stories from Threat Digest, delivered once a week.