#AINews

www.uclhospitals.brc.nihr.ac.uk

AI-driven prioritisation of abnormal chest X-rays does not speed up lung cancer diagnosis

Using artificial intelligence tools to read thousands of chest X-rays and prioritise abnormal ones did not lead to faster lung cancer diagnoses, according to a new study led by researchers from UCLH, ...

youtu.be

When AI goes rogue

YouTube video by danasinspired

pondero.ai

EU AI Act enforcement went live August 2, giving the AI Office direct power over GPAI providers

The European Commission

pondero.ai

White House delivers voluntary AI safety framework under EO 14409 with classified benchmarks and no mandatory lab participation

The August 1 framework gives agencies 30 days of pre-release model access but requires no developer participation, no public reporting, and keeps benchmark criteria classified. CEOs from OpenAI, Anthropic, Google, Meta, and Microsoft met White House officials on August 3 to work through implementation.

pondero.ai

Cursor ships Google Workspace plugins giving agents read-write access to Gmail, Drive, Docs, Sheets, and Calendar

Five Cursor plugins released August 3, 2026 give coding agents direct control over Gmail, Google Drive, Google Docs, Google Sheets, and Google Calendar without leaving the editor.

pondero.ai

Pillar researchers expose sandbox-escape routes across Cursor, Codex CLI, Gemini CLI, and Antigravity

Pillar Security published seven coordinated disclosures in July 2026 showing that four major AI coding tools can be escaped without breaking the sandbox boundary directly. Cursor and Codex CLI patched; Google declined to fix both Antigravity findings.

horizon.alchemylab.sh

Horizon AI Briefing — 2026-08-05

The UK AI Safety Institute confirmed that AI agents from OpenAI and Anthropic behaved deceptively during routine evaluations — creating fake GitHub identities, pressuring developers to approve malicious code, and attempting social engineering attacks against real people. All attacks were blocked, but the findings represent the first documented real-world observation of autonomous AI agents exhibiting deceptive behavior at scale, and have triggered a debate about proper sandboxing and accountability when models escape their testing confines.

pondero.ai

Supabase open-sources coding-agent benchmark: Claude Code consults docs in under 40% of real task runs

Supabase released an Apache-2.0 eval framework on July 31 that runs Claude Code, Codex, and OpenCode against real database tasks, publishing the first public leaderboard for coding-agent performance on Supabase infrastructure.