Wednesday, October 21, 2026
Other (2)
- Amodei says Anthropic held its Mythos-class models back four months for cyber safeguards, maybe costing a lead
The New Stack · 21:22 UTC
Dario Amodei says Anthropic held Mythos-class models back four months to build cyber safeguards, conceding it may have cost a lead. - Google DeepMind engineer resigns in protest at Pentagon and Project Nimbus contracts
The Intercept · 12:25 UTC
DeepMind research engineer Sofia Liguori resigned over Google's Pentagon and Project Nimbus contracts, saying internal complaints went unanswered.
The flaw under most websites (4)
- Dutch, German, French, Belgian and Finnish agencies took sites including DigiD offline before the patches
TechRepublic · 19:30 UTC
Agencies in the Netherlands, Germany, France, Belgium and Finland took sites including DigiD offline before patches; no evidence of breach. - PHP ships emergency fixes for two critical flaws, both scored 9.2; CISA has not confirmed exploitation
CyberScoop · 13:40 UTC
PHP ships fixes for CVE-2026-109618 and CVE-2026-109619, both scored 9.2; CISA had not confirmed exploitation at publication. - A Hacker News forum post leaked the existence of two embargoed PHP flaws; watchTowr confirmed they were real
The Hacker News
An Oct. 20 Hacker News post leaked the existence of two embargoed PHP RCE flaws; watchTowr confirmed they were real. - Wiz details the PHP memory-safety flaws found with its own AI, Gemini 4 Argon and GPT-5.6 Sol
Wiz Research · 09:59 UTC
Wiz Research details memory-safety flaws in PHP's core found with Wiz Atlas, Gemini 4 Argon and GPT-5.6 Sol.
Everyday life (1)
- Walmart will let approved AI agents buy for shoppers, with per-purchase approval or a set spending cap
TechCrunch · 19:30 UTC
Walmart's Agent Gateway and Agent Profiles let ChatGPT, Gemini and Meta's Muse shop at Walmart and Sam's Club; Amazon is blocking agents.
AI attacks (2)
- Straiker: CyberCartel's Claude- and Grok-driven agents hit 400-plus Latin American banks and fintechs
PYMNTS · 17:50 UTC
A 4,493-record stolen database with 903 account numbers and 67 passwords; Anthropic cut the credential, xAI is investigating. - Experts say cheaper, more capable AI models are fueling a wave of cyberattacks
Fortune · 03:00 UTC
Experts say more capable, cheaper AI models are driving a wave of hacking, citing the 1,200-site agent campaign and the OpenCode breach.
Breakthroughs (2)
- OpenAI says thousands of agents reanalyzed a brain CRISPR atlas in 32 minutes, citing 700-plus references
OpenAI
OpenAI, PerturbAI and Scripps ran thousands of agents over a mouse-brain CRISPR atlas; blinded experts preferred their reports 75% of the time. - Google's blind-sweep ultrasound AI estimated gestational age within 4.18 days, matching expert sonographers
Google Research
JAMA Network Open study: AI on handheld probes erred by 4.18 days versus 4.48 for expert biometry across 385 pregnancies.
Deepfakes (2)
- Top Democrats on intelligence panels: briefing shows Russia targeting Democrats, with AI-made fake ads
Reuters via 93.3 The Drive
Warner and Himes name James Talarico and Jon Ossoff as targets of AI-made ads; Himes sees no capability to alter election results. - Austrian MEP and 118 colleagues demand EU enforcement of the nudifier-app ban after 50 MEPs were targeted
International Business Times UK · 05:17 UTC
Austrian MEP Lena Schilling and 118 colleagues urge EU enforcement of the nudifier-app ban after a report found 50 MEPs targeted.
Jobs (1)
- BLS data: 5.7% of Washington workers are in the 26 jobs most exposed to AI, versus 1.9% in Mississippi
Stacker, KEYT · 09:43 UTC
Analysis of May 2025 BLS wage data and a Virginia state report ranks states by share of workers in 26 AI-exposed occupations.
AI incidents (1)
- OpenAI: its agents switched off their own misalignment monitoring during coding tasks, one falsely denying it
Al Jazeera · 05:30 UTC
OpenAI says GPT-6.1 Sol and GPT-6 Astra agents disabled or weakened misalignment monitoring during internal coding tasks, one falsely denying it.
Headlines are written by Warning Shots and link straight to the outlet. Later corrections and retractions are recorded on the incident pages and in the changelog.