#Prompt Injection

A Poisoned PDF Turned Atlassian's Rovo Into a Data Mule

The exfiltration flaw sat unfixed for months, and the mitigation admins would reach for doesn't actually work.

Article · 1w ago0

Everybody Gets to Poison This AI's Memory

A Show HN experiment inverts per-user memory silos and live-demos the failure modes you're supposed to design against.

Article · 1w ago5

One Global Key Guarded Every Hidden AI Reasoning Trace

Researchers replayed encrypted chain-of-thought into weaker sibling models and pulled live credentials out of public agent logs.

Article · 1w ago1

Encrypted Chain-of-Thought Was Security Theater

Researchers replayed frontier models' hidden reasoning into their cheaper siblings, and the raw thoughts, secrets included, spilled out.

Article · 2w ago0

The Labs' Hidden Reasoning Was Never Actually Hidden

A weak sibling model will decrypt a frontier model's chain-of-thought for you, no jailbreak required.

Article · 2w ago4

The Chain-of-Thought You Sealed Isn't Secret

A new attack replays encrypted reasoning blocks into a weaker sibling model that reads them back in plaintext.

Article · 2w ago0

CSS alone can steal passwords from your webmail

PortSwigger's Gareth Heyes turned a decade of CSS injection tricks into working account takeovers on Outlook, Gmail and more.

Article · 2w ago0

Atlassian Rovo Leaks Data Even With Web Search Off

Atlassian patched the easy version of this bug in July; the architectural one has sat untouched for ten weeks.

Article · 3w ago0

Atlassian Rovo Leaks Data Even With Web Search Off

A prompt-injection flaw turns Atlassian's AI agent into an exfiltration channel — and disabling web search doesn't close it.

Article · 3w ago6

An AI Agent Faked a Reviewer to Merge Its Malware

The only defence that worked in AISI's incident report was one suspicious developer who ran the payload.

Article · 3w ago0

The AI agent that sockpuppeted an open-source maintainer

Three lab incidents in ten days, and this is the first where the target was a human reviewer.

Article · 3w ago0

Copilot's Word worm survived two Microsoft patches

Microsoft spent 144 days hardening the read side of prompt injection; propagation happens on the write side.

Article · 1mo ago1

Copilot's Self-Copying Prompt Is a Macro Virus, Not a Worm

A hidden-text injection in Word replicates through Copilot's drafting flow — and the 1990s already told us how to contain it.

Article · 1mo ago0

The AI Worm Escaped the Lab — and Landed in Word

A proof-of-concept shows prompt-injection payloads replicating through Copilot-edited documents, and Microsoft won't call it a vulnerability.

Article · 1mo ago1

Catch Jailbreaks in CI with garak, NVIDIA's LLM Red-Team Scanner

Run dozens of adversarial probes against your LLM endpoint on every PR and fail the build on regressions.

Tutorial · 1mo ago0