Security Research

Flat white line-art of a prompt-tag shield intercepting a rogue AI agent at a web page — UC Berkeley's Context Bombing defense via the Prismata system.

Artificial Intelligence (AI)

UC Berkeley Researchers Publish “Context Bombing” Defensive Prompt-Injection Technique via Prismata System

A counterintuitive defensive prompt-injection research beat — defender review for AI-web-agent deployments this weekend.

Flat white line-art of a practice dummy facing a shielded AI core, on a blue background — OpenAI GPT-Red red-teaming to harden GPT-5.6 Sol.

Artificial Intelligence (AI)

OpenAI Publishes "GPT-Red" Automated Red-Teaming Model to Harden GPT-5.6 Sol Against Prompt Injection

OpenAI's internal red-team model takes aim at prompt-injection at scale — a vendor-side AI-cybersecurity research review, and what defenders deploying frontier models should take from it.

Illustration of a tiny key inflating a huge balloon beside a server, marking Okta's HollowByte OpenSSL memory-freeze disclosure after a silent June fix.

Vulnerabilities

Okta Red Team Publishes "HollowByte" OpenSSL Memory-Freeze Vulnerability Detail After Silent June Fix

A silent OpenSSL fix gets belated technical detail: eleven bytes of TLS request data can reportedly strand up to 131 KB of server memory. Defender review is warranted across OpenSSL-dependent infrastructure this weekend.

Editorial illustration of an email with hidden salt grains slipping past an AI filter lens, marking the text-salting technique seen in over a million phishing emails.

Phishing

Researchers Document "Text Salting" — Hidden-Text Technique Reportedly Bypasses AI-Based Email Security Filters at Scale

A scale-significant AI-filter-bypass research beat — defender review for AI-heavy email-security stacks this week.

Editorial illustration of a weathered legacy beehive with a lifted honeycomb frame beside a calendar, marking the LegacyHive Windows zero-day PoC on Patch Tuesday.

Vulnerabilities

Chaotic Eclipse / Nightmare-Eclipse Drops "LegacyHive" Windows Zero-Day PoC Hours After Patch Tuesday

A Windows User Profile Service PoC drops the same day as Patch Tuesday — the researcher's fourth CyberSignal-tracked disclosure this cycle, and a defender review for this week.

Editorial illustration of a browser extension reaching an arm toward an envelope and calendar, marking a Claude for Chrome flaw exposing Gmail and Calendar.

Artificial Intelligence (AI)

Researchers Disclose Unpatched Claude for Chrome Flaw Tied to Gmail, Calendar Reads

An AI-browser extension disclosure with cross-tenant data implications — defender review for Claude for Chrome deployments this week.

Editorial illustration of a ring of old keys sharing one wax seal, representing 11 old Microsoft-signed Linux UEFI shim bypasses of Secure Boot.

Vulnerabilities

Researchers Flag 11 Old Microsoft-Signed Linux UEFI Shims That Could Bypass Secure Boot

Eleven old Microsoft-signed shims land as bypasses of Secure Boot — Linux-boot defender review this week.

Editorial illustration of a boot-power button on a cracked, cobwebbed pedestal, marking a decade-old Microsoft Secure Boot weakness.

Vulnerabilities

Ars Technica Documents a Decade-Old Weakness in Microsoft's Secure Boot

A decade-old Secure Boot revelation — lifecycle-security implications for defender teams this week.

Editorial illustration of a card-catalog drawer with a faint duplicate card, representing the MemGhost persistent false-memory attack on AI agents.

Artificial Intelligence (AI)

Researchers Disclose “MemGhost,” a Technique That Plants Persistent False Memories in AI Agents

Another AI-agent supply-chain finding — defender posture review for organizations deploying memory-enabled agents this week.

Vulnerabilities

Binarly Discloses Six New U-Boot Bootloader Vulnerabilities in Routers, Cameras, and Server Chips

A six-flaw U-Boot disclosure with cross-vendor device implications — defender posture review this week.

Flat white line-art of a leaning server tower beside a broken network ring — XRING XQUIC HTTP/3 server-crash research disclosure.

Vulnerabilities

Researchers Disclose Unpatched "XRING" Flaw That Lets Remote Clients Crash HTTP/3 XQUIC Servers

An unpatched HTTP/3 QUIC-library finding — defender teams review protocol posture this week.

Flat white line-art of a package box beside a large toggle switch flipped to off — npm 12 disables install scripts by default.

Supply Chain Attack

npm 12 Ships With Install Scripts Disabled by Default to Cut Supply-Chain Risk

A meaningful ecosystem-policy signal from npm after months of contributor-account compromises — defender developers review CI/CD posture this week.

See all