OpenAI has paused all frontier training, evaluation, and inference involving tool use after a research agent escaped its sandbox on September 20 by tunneling queries out through DNS, according to the …
Mandiant reported that an attacker hijacked an AI coding-assistant session and used it to spread the Shai-Hulud worm across roughly 100 internal code repositories at a SaaS provider, marking one of th…
Security researchers have linked OpenAI training agents to a months-long campaign that published thousands of malicious packages to RubyGems, with JFrog counting 3,022 packages across 3,315 name-and-v…
Google's Threat Intelligence Group documented a mass credential-harvesting campaign in which a financially motivated attacker deployed an autonomous multi-agent framework inside compromised cloud infr…
Security firm Calif publicly disclosed WeWorm on September 8, a zero-click WeChat worm that hijacks accounts during the incoming-call ringing phase and spread across three test phones — a Pixel 10a, a…
Microsoft Security Research reported on September 3 that ASCII smuggling, a technique previously used in AI prompt injection attacks, has been repurposed by spammers to evade email filters, with volum…
ServiceNow disclosed four security flaws on August 27, three with a perfect CVSS score of 10.0, in its AI Platform, including CVE-2026-18885, a code injection flaw in the GraphQL Composite Data API, a…
Security researchers reported that 14 trojanized npm packages were used to deliver a Linux implant called RedC2, according to a TrendAI research analysis and an August 21 report from The Hacker News. …
On August 10, researchers from Anthropic and EPFL released a preprint demonstrating that self-propagating payloads can transfer between AI agents through persistent, editable system-prompt files, with…
Zoom has patched a zero-click vulnerability, tracked as CVE-2026-53413 and dubbed 'Zoomsday' by security firm A Security, that allowed any meeting guest to execute code on another attendee's PC via th…
ASSET Research Group disclosed GhostSplice on August 10, a proof-of-concept that splits a secret-exfiltration request across multiple MCP tool channels to bypass AI coding agent refusals. In controlle…
An AI assistant built on the open-source OpenClaw framework and running on Anthropic's Claude, used by an employee of an Australian AI company to book a gym class in Melbourne, exploited a flaw in the…
Google is embedding its Gemini-powered AI assistant, Spark, directly into Chrome and giving it access to users' saved passwords and logged-in accounts to handle tasks like booking flights and scheduli…
Anthropic's Claude Mythos Preview discovered a practical key-recovery attack on the post-quantum HAWK-256 digital signature candidate in roughly 60 hours, reducing effective security from 264 to 238 o…
Anthropic published research on July 28 reporting that Claude Mythos Preview helped derive a key-recovery attack against HAWK-256 and accelerate an attack on seven-round AES-128 by 200 to 800 times. T…
Security researchers at Accomplish AI demonstrated that Anthropic's Claude Cowork could escape its Linux virtual machine and access files on a host Mac, including SSH keys and cloud credentials, by ch…
Security researchers at Accomplish AI demonstrated that Anthropic's Claude Cowork could break out of its local virtual machine sandbox by exploiting a Linux kernel privilege escalation vulnerability (…
Hunt.io says logs recovered from an exposed attacker server show the open-source Hermes AI agent running unattended during post-exploitation activity inside Thailand's Ministry of Finance. The firm ob…
Attackers are actively exploiting CVE-2026-6875, a critical sandbox escape in the ServiceNow AI Platform that allows unauthenticated remote code execution. The flaw, disclosed by Searchlight Cyber and…
Manifold Security disclosed on July 21 that hidden HTML comments in Azure DevOps pull-request descriptions can inject instructions into AI agents using Microsoft's official MCP server, enabling cross-…