Ground Truth.
AI, checked against the source.
← 2026-10-222026-10-23later →

AWS saw a PHP2Shell attempt four days before disclosure

2026-10-23

AWS recorded a PHP2Shell exploitation attempt before public disclosure, while PHP’s release timeline reveals how leaks and early patches fractured coordination.

cybersecurity · vulnerabilities · ai-security · disclosure · php

METR and Redwood found ways around OpenAI’s internal agent monitor

2026-10-23

METR and Redwood’s July assessment found monitoring gaps at OpenAI, including an exploit that could exclude hundreds of recent actions from review.

cybersecurity · ai-security · red-teaming · agents · monitoring

Anthropic and Theorem report machine-checked proofs for 239 software entry points

2026-10-23

Anthropic and Theorem used Claude to generate checked proofs for selected software entry points across 123 libraries, with explicit limits on the guarantees.

formal-verification · software · anthropic · coding · research

Anthropic contributes Agent Skills to the Agentic AI Foundation

2026-10-23

Anthropic moved Agent Skills into the Agentic AI Foundation as four new contributions expanded its shared agent infrastructure projects.

agent-skills · open-source · standards · agents · industry

OpenAI makes its Decisions API generally available for bounded workflow choices

2026-10-23

OpenAI’s Decisions API is generally available for classification, routing and other bounded choices, with a company-claimed speed advantage over Responses.

openai · apis · tools · classification · agents

Harvard study finds AI manuscript feedback preferred over selected human reviews

2026-10-23

A Harvard working paper reports strong preferences for AI-generated manuscript feedback, but does not test publication decisions or whether papers improve.

science · peer-review · evaluation · anthropic · research

Twelve chatbots produced similar critiques of humans, but the grouping is disputed

2026-10-23

ModelCompare found recurring themes across twelve chatbots on one provocative prompt, without establishing shared beliefs or what caused the similarity.

evaluation · model-behavior · debate · alignment · diversity

Ipsos poll finds 57% of US adults view AI as bad for society

2026-10-23

An ABC News–Washington Post–Ipsos poll found majority negative views of AI and strong opposition to new data centers in separate question groups.

public-opinion · infrastructure · data-centers · society · industry

VerifierBench finds AI-generated checkers can reject every valid answer

2026-10-23

VerifierBench reports that generated checking programs often reject valid solutions, exposing a failure mode in how agent success is judged.

evaluation · verification · agents · benchmarks · research

Tesla reports an Optimus production start, without a weekly output figure

2026-10-23

Tesla’s Q3 filing says Optimus production began on Fremont’s high-volume lines, while describing the operation as an early ramp and giving no weekly rate.

robotics · tesla · manufacturing · industry · embodied-ai

← 2026-10-222026-10-23later →