Humans are reviewing Copilot users’ bizarre and abusive image-editing requests

A reminder for anyone who thinks their AI chats are private: A human may see what they upload. A report from 404 Media this week describes contractors reviewing Copilot users’ photos, editing requests, and the images Copilot produces. Some of what they see is disturbing. The third-party companies...

Invisible Ink Threats: Adversarial Goals behind Legitimate Tasks in Computer-Use Agents

Computer-use agents CUAs, which empower large language models to autonomously operate operating systems and the web, are increasingly vulnerable to indirect prompt injection attacks. A widely adopted defense is the human-in-the-loop paradigm, in which the agent pauses for explicit user confirmati...

The Prover Is the Judge: Verified Security Software from AI Coding Agents in Ada/SPARK

AI coding agents produce code faster than humans can review it. In our approach, the prover is the judge of whether the code is correct. Under a verifier-driven loop, AI agents wrote and verified bare-metal security software in Ada/SPARK spanning classical and post-quantum cryptography, TLS 1.3,...

PYSEC-2026-2901 PraisonAI: Unauthenticated Allow-List Manipulation Bypasses Agent Tool Approval Safety Controls

Summary The gateway's /api/approval/allow-list endpoint permits unauthenticated modification of the tool approval allowlist when no authtoken is configured the default. By adding dangerous tool names e.g., shellexec, filewrite to the allowlist, an attacker can cause the ExecApprovalManager to...

MAL-2026-6305 Malicious code in @frostnode/waitfor (npm)

@frostnode/waitfor malicious versions 0.9.0, 0.10.3, 0.10.4, and 0.10.5, published by [email protected] is a trojanized npm package belonging to the wshu.net credential-stealer campaign. The campaign published trojanized look-alike utility packages across 12+ scopes whose publisher accoun...

Malicious code in @frostnode/waitfor (npm)

@frostnode/waitfor malicious versions 0.9.0, 0.10.3, 0.10.4, and 0.10.5, published by [email protected] is a trojanized npm package belonging to the wshu.net credential-stealer campaign. The campaign published trojanized look-alike utility packages across 12+ scopes whose publisher accoun...

CyBOKClaw: Human-In-The-Loop CyBOK Mapping for Cybersecurity Curriculum

This paper presents CyBOKClaw, an interpretable human-in-the-loop retrieval framework for mapping cybersecurity keywords or phrases KWoPs to the Cyber Security Body of Knowledge CyBOK. Rather than treating the task as strict exact classification, the framework is designed as a top-k candidate...

AI Is Here to Replace Nuclear Treaties. Scared Yet?

The last major nuclear arms treaty between the US and Russia just expired. Some experts believe a combination of satellite surveillance, AI, and human reviewers can take its place. Others, not so much...

What AI Reveals About Web Applications— and Why It Matters

Before an attacker ever sends a payload, they’ve already done the work of understanding how your environment is built. They look at your login flows, your JavaScript files, your error messages, your API documentation, your GitHub repos. These are all clues that help them understand how your syste...

Apple: Image-Detection Backdoor ‘Narrow’ in Scope

Apple provided additional design and security details this week about the planned rollout of a feature aimed at detecting child sexual abuse material CSAM images stored in iCloud Photos. Privacy groups like the Electronic Frontier Foundation warned that the process of flagging CSAM images...

Apple Changes the Way It Listens to Your Siri Recordings Following Privacy Concerns

Apple today announced some major changes to its controversial 'Siri audio grading program' following criticism for employing humans to listen to audio recordings of users collected via its voice-controlled Siri personal assistant without their knowledge or consent. The move came a month after The...

Google Makes it Tough for Rogue App Developers Get Back on Android Play Store

Even after Google's security oversight over its already-huge Android ecosystem has evolved over the years, malware apps still keep coming back to Google Play Store. Sometimes just reposting an already detected malware app from a newly created Play Store account, or using other developers' existin...

Uber: Users can falsely declare their own Uber account info on the monthly billing application

When applying for monthly billing, statistics and data regarding account usage that was fetched and subsequently submitted could be tampered with. This data would be reviewed by a human making a credit decision, potentially influencing their decision inappropriately in favor of the applicant. Whe...