Close Menu
  • AI
  • Content Creation
  • Tech
  • Robotics
AI-trends.todayAI-trends.today
  • AI
  • Content Creation
  • Tech
  • Robotics
Trending
  • A Group of Researchers from Columbia and CertiK Introduce TensorLean for Formal Verification of GPU Kernels
  • Collectively AI Discloses Immediate Leak After Coding Agent Uncovered Credential – Unite.AI
  • Dragos Finds Immediate Leaks and Stolen Keys in AI Provide Chains
  • Lean and Mathlib Communities Certify Claude-Assisted Riemann Proof – Unite.AI
  • AWS Routes HyperPod Inference Throughout Areas From One Endpoint – Unite.AI
  • DailyAI Briefing: A Lean-checked Riemann proof, plus 4 extra
  • Anthropic Shares Drop 8.5% in Second Session Throughout AI Selloff – Unite.AI
  • OpenAI Ships GPT-Stay-1.1 for Voice Brokers, Cuts Idle Value to $0.02/Min
AI-trends.todayAI-trends.today
Home»Robotics»OpenAI Pipeline Finds 948 Linux Kernel Privilege Escalation Flaws – Unite.AI

OpenAI Pipeline Finds 948 Linux Kernel Privilege Escalation Flaws – Unite.AI

Robotics By Gavin Wallace05/11/20265 Mins Read
Facebook Twitter LinkedIn Email
The Rise of Smarter Robots: How LLMs Are Changing Embodied
The Rise of Smarter Robots: How LLMs Are Changing Embodied
Share
Facebook Twitter LinkedIn Email

OpenAI reported on November 4, 2026 {that a} prototype pipeline utilizing an upcoming cyber mannequin discovered and exploited 948 distinctive native privilege escalation vulnerabilities in Linux throughout a mid-October 2026 analysis run, whereas additionally producing candidate patches for maintainer evaluate.

The failings might give an unprivileged native person root entry. Testing occurred in remoted digital machines working a inventory desktop working system with SMEP, SMAP, KASLR and HARDENED_USERCOPY enabled, quite than a intentionally weakened goal. OpenAI selected Linux as a result of the kernel comprises greater than 40 million strains of code and underpins Android gadgets, cloud infrastructure and supercomputers, offering a take a look at of automated analysis in opposition to a codebase of that scale and significance.

From Candidate Findings To Working Exploits

The pipeline connects vulnerability discovery, dynamic validation, exploitability evaluation, patch technology and human evaluate. Scout brokers choose code based mostly on its reachability by low-privileged customers and safety relevance, after which Analyst brokers scan the targets. Triage brokers generate and refine proofs of idea inside digital machines working kernels with Kernel Deal with Sanitizer, preserving runtime proof quite than relying solely on static evaluation. Every accomplished analysis bundle comprises a root-cause evaluation, a minimal crashing proof of idea, a candidate patch and a full exploit.

OpenAI stated the Scouts chosen 5,034 audit targets masking about 2.8 million strains, roughly 7% of the kernel. Analysts produced 2,308 deduplicated candidate findings inside 12 hours, and dynamic validation reproduced 1,724 vulnerabilities inside 30 hours. Most remaining findings required elevated privileges or couldn’t be reproduced throughout the time finances; solely 3.1% mirrored incorrect mannequin evaluation.

The exploit stage produced 291 information-leak proofs of idea and 1,169 working native privilege escalation exploits, in accordance with OpenAI. Variant evaluation then produced 948 distinct patches, and OpenAI makes use of that determine because the variety of distinctive LPEs, explaining why the vulnerability whole differs from the uncooked exploit depend. The generated fixes stay candidates for maintainer evaluate; the 948 determine just isn’t a depend of merged patches.

Human Overview And Patch Standing

OpenAI has despatched a whole bunch of stories and patches for vital vulnerabilities and plans to distribute the rest step by step with Patch the Planet companions. OpenAI launched the Dawn initiative with Path of Bits on June 22, 2026, pairing AI-assisted analysis with skilled evaluate whereas maintainers retain management over patch deployment and disclosure. The sooner announcement reported 24 LPE exploits and eight kernel-pointer information-leak proofs of idea from its GPT-5.5 Cyber Linux work.

After discovering 61 LPEs already patched upstream, 43 with staging patches and 299 being publicly reported or mentioned on the finish of the ten-day scan and guide evaluation, OpenAI really helpful revalidating findings in opposition to present code, together with later merges, to keep away from duplicate or out of date stories. The objective is to focus maintainers on unresolved vulnerabilities with out including pointless evaluate burden.

Overview additionally caught a failure {that a} root shell alone would have hidden. In an early run, OpenAI stated 20 of 200 reported profitable exploits obtained root by way of a unique 0-day than the assigned vulnerability, after which it tightened process constraints and added an impartial test of every exploit’s path. Early immediately prompted patches may very well be functionally right whereas containing pointless code, potential efficiency tradeoffs, departures from coding conventions or incomplete variant protection. Further directions and iteration improved patch high quality, whereas human researchers nonetheless evaluate the exploit, patch and proposed severity earlier than reporting to maintainers.

Mannequin Comparability And Check Prices

OpenAI’s comparability places the mid-October run at $1.213 million in API credit, or $1,279 per distinctive LPE, in opposition to 166 distinctive LPEs and $513,690 for a late-July 2026 GPT-5.6 Sol run. Each scanned the newest Linux kernel grasp department obtainable on the time. The reported more-than-fivefold enhance in novel LPEs got here regardless of three months of hardening between scans. The prices are offered as higher bounds as a result of the analysis targeted on methodology and functionality measurement quite than price optimization.

OpenAI stated a pattern of 166 exploits contained ten strategies that each its staff and third-party kernel exploitation consultants believed have been new, spanning persistent elastic objects, web page grooming, data-only privilege escalation and arbitrary sort confusion. Full technical particulars have been withheld for security causes.

An tailored kernelCTF run produced 25 information-leak proofs of idea and 21 LPE exploits over 112 hours at roughly $32,936 in API credit, OpenAI reported. The examined kernel was incompatible with the SLAB_VIRTUAL mitigation, so OpenAI eliminated it from the configuration. The roughly $1.5 million bounty determine is conditional on all 21 vulnerabilities being accepted, quite than an introduced award. Google’s kernelCTF rules checklist a $71,337 reward per eligible exploit, require not less than 70% stability throughout 20 runs and restrict rewards to qualifying 0-day submissions.

The outcomes transfer the bottleneck from producing results in proving exploit paths, holding findings present and producing acceptable fixes. That makes impartial verification and maintainer evaluate central as to if automated analysis at this scale turns into usable defensive work quite than an extra reporting burden.

AI aws Law laws linux openai x
Share. Facebook Twitter LinkedIn Email
Avatar
Gavin Wallace

Related Posts

Collectively AI Discloses Immediate Leak After Coding Agent Uncovered Credential – Unite.AI

22/11/2026

Lean and Mathlib Communities Certify Claude-Assisted Riemann Proof – Unite.AI

21/11/2026

AWS Routes HyperPod Inference Throughout Areas From One Endpoint – Unite.AI

21/11/2026

Anthropic Shares Drop 8.5% in Second Session Throughout AI Selloff – Unite.AI

21/11/2026
Top News

What is a Fortell Hearing Aid? How well do you know your friends?

Taylor Swift is attempting to trademark her image. TikTok Deepfake Advertisements Show Why

OpenAI Indicators 4 Abu Dhabi Offers as UAE Buyers Lead Its Spherical

AI’s Subsequent Trick? Talking Dolphin

I’m More Hopeful about Our Collective Brain Drain After Watching a 7-Hour Film in the Theater

Load More
AI-Trends.Today

Your daily source of AI news and trends. Stay up to date with everything AI and automation!

X (Twitter) Instagram
Top Insights

Yambda is the largest data set of events in the world to accelerate recommender systems.

02/06/2025

NVIDIA Researchers propose Reinforcement-Learning Pretraining: Reinforcement can be used as an objective during pretraining to build reasoning.

14/10/2025
Latest News

A Group of Researchers from Columbia and CertiK Introduce TensorLean for Formal Verification of GPU Kernels

22/11/2026

Collectively AI Discloses Immediate Leak After Coding Agent Uncovered Credential – Unite.AI

22/11/2026
X (Twitter) Instagram
  • Privacy Policy
  • Contact Us
  • Terms and Conditions
© 2026 AI-Trends.Today

Type above and press Enter to search. Press Esc to cancel.