Anthropic releases Claude Fable 5.2 and restricted Mythos 5.2
API prices fall 20% to $8 and $40 per million tokens, while Mythos access stays limited to vetted organizations.
1
Stories per day · 30 days
3 RSS
Relatedopenai4anthropic2claude2linux2model-release2mythos2prompt-injection2agent-harness1ai-agents1ai-discovered-bugs1ai-security1benchmarks1codeql1commentary1daybreak1exploitation1fable1false-negatives1
30 stories, newest first · one card per story
Stories tagged #vulnerability-discovery
OpenAI's GPT-6 Cyber finds 81 unknown high-severity Firefox bugs and 588 kernel flaws
OpenAI claims 10 of the model's exploitation techniques were novel; its dashboard shows under 25% of found flaws remediated.
1
Athena Project Discloses First 50 Silent Fixes
Athena discloses 50 silent fixes from 40,000 findings, partnering with CISA for responsible disclosure.
OpenAI Pipeline Generates 948 Linux Root Escalation Exploits
OpenAI's AI pipeline discovered and patched 948 Linux local privilege escalation exploits in 30 hours at $1.2 million cost, using an upcoming cyber model.
1
Google unveils Gemini 4 Carbon for flaw finding, exploitation tests, and patches
Selected defenders get first access; Google cites 88.2% on SEC-bench Pro and has not set a general-availability date.
Experts warn AI-found bugs may cause new disruptions
The available feed text is only the headline and the word Comments, with no supporting detail.
LLM vulnerability harnesses can drop real bugs before the model runs
Authors mapped 13 drop points across 12 harnesses; six discarded one file unseen, five without notice.
Study measures how harness and context change LLM vulnerability discovery
No single setup covered all 265 confirmed root causes, and CodeQL context helped only the smallest of four models.
Where Do the Tokens Go? Understanding and Reducing Costs in LLM Agents for Vulnerability Discovery
A 200-trace CyberGym study finds reasoning dominates token cost, and AVRI lowers it.
Anthropic Expands Claude Access for Vetted Cyber Teams as Glasswing Finds 129,000 Flaws
Anthropic expands vetted Claude cyber access after Glasswing verified 129,000 software flaws.
8
Google's PageBreak AI Agent Finds 500 Flaws in Its Web Apps
Google's PageBreak AI agent reportedly found 500 flaws in the company's own web applications.
Google PageBreak AI Agent Finds Over 500 XSS Vulnerabilities Across Its Web Applications
Google's PageBreak AI agent validated over 500 XSS flaws in its own web apps using exploit-backed deterministic validation with near-zero false positives.
Insights from the 2026 Microsoft Digital Defense Report
Microsoft's 2026 defense report says attackers use AI while agent and identity security grows harder.
4
Google Rolls Out Gemini 4 Argon to Trusted Cyber Defenders, Plans Guardrail-Free Version
Google DeepMind's Gemini 4 Argon frontier model rolls out to trusted cyber defenders, with autonomous vulnerability discovery and a planned guardrail-free version.
3
Google says Gemini 4 Argon can find and patch critical software flaws
Google launched frontier model Gemini 4 Argon, claiming it can find and patch critical software flaws without human help.
4
Google to critical infra orgs: Our AI scanners won't be evil, promise
Google and Wiz launch Scan for Good, using Gemini 3.8 Flash Cyber and Red Agent to autonomously find critical vulnerabilities in hospitals, transit, and municipalities.
State-Aware Fuzzing of JavaScript Engines with LLM-Guided Instrumentation
StateLens uses LLM-guided instrumentation to fuzz JavaScript engines and found 68 bugs.
MobileCybench: Evaluating Agent Vulnerability Discovery via Executable Probes
MobileCybench uses 495 executable probes to score AI agents finding flaws in 13 Android apps.
Security through obscurity is dead, and AI delivered the fatal blow
AI agents are killing security through obscurity, accelerating vulnerability discovery and patch-gap exploitation, experts warn, with OT/ICS systems most at risk.
3
Google Open-Sources Mantis: A Modular Skills Toolkit That Lets Coding Agents Find, Reproduce and Patch Vulnerabilities
Google open-sourced Mantis, an Apache-2.0 modular skills toolkit that lets AI coding agents find, reproduce, and patch vulnerabilities with sandboxed verification.
4
Hunting Vulnerabilities Using Frontier Models
Okta used frontier AI models GPT-5.5 Cyber and Mythos via OpenAI and Anthropic programs to scan millions of code lines for vulnerabilities.
Most of the bugs Claude Mythos found have never been checked by a human
Echo's analysis found only 1,900 of 23,019 Claude Mythos-found vulnerabilities were externally reviewed, 90.8% held up, but the model overstated most severities.
How AI could make it harder for governments to use hacking tools
TechCrunch analysis argues AI-driven vulnerability discovery and exploitation may constrain government use of hacking tools and spyware.
AI vulnerability discovery scores the highest impact of 20 emerging risks
Gartner survey of 316 organizations ranks AI-driven vulnerability discovery as the top emerging risk, with tangible impact expected within roughly two years.