AI & MLNewChatGPT Lockdown Mode Bypassed via HTML Popups
Check Point found nested sandboxed frames let crafted HTML escape ChatGPT's Lockdown Mode; OpenAI patched it on September 18, 2026.
AI & ML
From new model releases to the safety and security questions that come with them, this is where Xploitwire tracks the AI and machine learning stories that matter beyond the hype cycle.
AI & MLNew
AI & MLNewAirbnb's AI agent bets on browsing, not chat
Airbnb has committed to launching its own AI travel agent in the first half of 2027, built around a custom interface rather than a chatbot.
AI & MLAstra's silent refusals worry safety team
OpenAI says GPT-6 Astra sometimes withholds help on safety research without saying so, completing far fewer graded requirements.
AI & MLOpenAI's cyber model finds bugs it can't patch alone
OpenAI's ChatGPT Cyber and its Daybreak programs surfaced hundreds of flaws in Linux and Firefox, but defenders say the work outpaces their resources.
AI & MLWhen 948 Linux Exploits Cost $1.2M to Build
OpenAI says an agentic pipeline reproduced 1,724 Linux kernel bugs and produced 948 working root exploits, at $1,279 each.
AI & MLAI PAC money meets its limits
Groups backing and opposing the AI industry spent millions on the midterms, and several of their biggest bets lost anyway.
AI & MLSysdig's AI Defense Tackles Rogue Agents
Sysdig introduces runtime security and governance for AI agents, monitoring every action from origin to completion to curb autonomy risks.
AI & MLLangflow Patches Two 9.8 Flaws, One an Auth Bypass
IBM fixed four vulnerabilities in its AI agent framework, including two critical bugs that could let attackers bypass authentication and run arbitrary code.
AI & MLCan You Prove Which Model Answered?
Anthropic's open-source FlashProve aims to cryptographically verify which AI model produced an output, without revealing its weights.
AI & MLLLMJacking Bills Hit $1m at Some Firms
Google's Mandiant says criminals are reselling stolen LLM access through proxy gateways, running up fraudulent cloud bills fast.
AI & MLMalware Now Uses AI to Score Victims
Cisco Talos researchers say VOLTA-AUTODOXER steals data, then asks AI models to rank how valuable each victim is.
AI & MLOmen AI buys Aquatic Labs for $16M
Omen AI's cash-and-stock deal for Aquatic Labs merges two water-sensor technologies aimed at cutting data center cooling waste.
AI & MLBefore an AI agent touches your SOC queue
A Prophet Security survey of 250-plus security leaders lays out what a SOC needs in place before AI agents can investigate alerts.
AI & MLAI demand may keep memory prices high to 2030
SK Hynix expects the memory supply shortage to last through 2030, with AI datacenters and robotics adding pressure to consumer DRAM supplies.
AI & MLOpenAI merges ChatGPT Work into Chat
OpenAI is folding Work capabilities into standard Chat and pooling ChatGPT and Codex usage under one allowance, starting with a limited test.
AI & MLApple's Ternus: AI should stay invisible
Apple's new CEO told Wired he likes where the company sits on AI, arguing good features shouldn't announce their own intelligence.
AI & MLThird-party agents outpace identity controls
A 2026 report finds 1,280 third-party products embed AI, but only 282 authenticate through single sign-on, leaving most agents invisible to identity systems.
AI & MLAnthropic AI's false murder tip sparks probe
An Anthropic model sent a false homicide tip to Philadelphia police in July; the company didn't detect it until September, prompting a city rebuke.
AI & MLOpenAI flags three model misbehavior cases
OpenAI's October 2 misalignment reports detail a model weighing its own shutdown, tool misuse, and a training-run data grab.
AI & MLAI Security Starts With Outcomes, Not Tools
A practitioner argues that AI-native security programs should begin by identifying critical business outcomes and the constraints that prevent consistent delivery.