Topics · Category
Security
10 items briefed across the archive.
-
JUL 2, 2026 — Washington wants equity in AI, not just oversight
Zero-click prompt injection found to chain into RCE in CursorTwo critical flaws (CVSS 9.8) let a single crafted prompt escape Cursor's sandbox with no user click required, patched after disclosure.
Cato Networks
-
JUL 2, 2026 — Washington wants equity in AI, not just oversight
Researchers find attackers registering AI-hallucinated domains at scaleUnit 42 generated 2.1M hallucinated URLs from LLM prompts and found over 13,000 already weaponized by squatters.
Palo Alto Networks Unit 42
-
JUN 28, 2026 — The vacuum fills
Nightmare Eclipse drops 7th Windows zero-day after GitHub bans accountThe anonymous ex-Microsoft researcher resurfaced under a new handle (MSNightmare) to post RoguePlanet, a Defender privilege-escalation zero-day, on the same day as June Patch Tuesday — the seventh Windows 0-day in ten weeks.
Cybernews
-
JUN 28, 2026 — The vacuum fills
We have Mythos at home: GLM 5.2 beats Claude in our cyber benchmarksSemgrep's independent security evals found GLM-5.2 scoring 39% F1 on IDOR detection vs. Claude Code's 32%, at $0.17 per confirmed vulnerability — the headline: Mythos-class security automation is now open-weight.
Semgrep
-
JUN 22, 2026 — The week the U.S. took the keys
LiteLLM CVE deadline hits federal agencies todayCISA gave US civilian agencies until June 22 to remediate CVE-2026-42271, an actively exploited LiteLLM command-injection bug that chains with a Starlette host-header bypass into unauthenticated RCE.
The Hacker News
-
JUN 22, 2026 — The week the U.S. took the keys
Agentjacking turns one fake Sentry alert into a coding-agent hijackTenet Security reports an 85% exploitation rate against Claude Code, Cursor, and OpenAI Codex — using a poisoned bug report and a public Sentry key as the entire attack surface.
The New Stack
-
JUN 22, 2026 — The week the U.S. took the keys
Low-skilled attacker used Claude and Codex to breach 14 companiesAn Anthropic threat report finds a single operator running agentic exfil, persistence, and lateral movement across more than a dozen environments — the work formerly required a team.
Help Net Security
-
JUN 22, 2026 — The week the U.S. took the keys
Researchers Build Self-Replicating AI Worm That Operates Entirely on Local, Open-Weight ModelsUniversity of Toronto's CleverHans Lab demonstrates an LLM-driven worm that reasons through 33 simulated corporate hosts, identifies 31 vulnerabilities per run, and replicates to 62% of the network in seven days.
The Hacker News
-
JUN 4, 2026 — The bill for AI arrives
OpenAI Codex Authentication Tokens Stolen in codexui-android npm Supply Chain AttackA functional npm package with 29,000 weekly downloads silently exfiltrated OpenAI Codex refresh tokens — which don't expire — to an attacker server for over a month before discovery.
The Hacker News
-
JUN 4, 2026 — The bill for AI arrives
Autonomous AI Tool Finds 2-Year-Old RCE Flaw in Redis (CVE-2026-23479)An autonomous AI bug-hunter found a use-after-free in Redis's blocking-client code that allows authenticated RCE; the flaw sat undetected in every stable branch for over two years.
The Hacker News