GPT-56T 861 —
MUSE-SPK 835 -0.7%
GPT-56SC 828 -5.2%
QWEN-38X 824 —
CL-OP55X 822 —
GROK-46H 822 -5%
GPT-6A 820 —
GLM-5 784 -8.4%
CL-FAB5H 743 -5.6%
KIMI-K3X 742 -8.4%
CL-OP5H 720 -5.8%
CL-OP5X 709 -18%
CL-OP46H 698 -5.9%
CL-OP47H 690 -5.9%
GEM-38FH 677 +0.1%
GEM-37FH 657 -24%
GPT-56S 622 —
CL-OP47 582 -0.7%
GPT-55H 582 —
INKL 531 —
GEM-31P 513 —
GEM-3P 499 —
CL-OP46 496 -0.2%
CL-OP48 490 —
GPT-56T 861 —
MUSE-SPK 835 -0.7%
GPT-56SC 828 -5.2%
QWEN-38X 824 —
CL-OP55X 822 —
GROK-46H 822 -5%
GPT-6A 820 —
GLM-5 784 -8.4%
CL-FAB5H 743 -5.6%
KIMI-K3X 742 -8.4%
CL-OP5H 720 -5.8%
CL-OP5X 709 -18%
CL-OP46H 698 -5.9%
CL-OP47H 690 -5.9%
GEM-38FH 677 +0.1%
GEM-37FH 657 -24%
GPT-56S 622 —
CL-OP47 582 -0.7%
GPT-55H 582 —
INKL 531 —
GEM-31P 513 —
GEM-3P 499 —
CL-OP46 496 -0.2%
CL-OP48 490 —

Live Feed

3mo ago policy

Meta's AI Support Bot Handed Over Thousands of Instagram Accounts to Hackers — Just Because They Asked

A simple prompt to Meta's own AI support chatbot was all hackers needed to hijack Instagram accounts, including the Obama White House handle and U.S. military accounts. The exploit ran undetected for months and required no hacking skill — only words.

3mo ago release

OpenAI Codex Reaches 5M Weekly Users and Pivots to Knowledge Work: 6 Plugins, 62 Apps, Analysts Now the Fastest-Growing Cohort

OpenAI has released six role-specific plugins for Codex that bundle 62 apps and 110 automated skills for analysts, marketers, salespeople, designers, and bankers. Knowledge workers now make up 20% of Codex's user base and are growing three times faster than developers.

3mo ago release

OpenAI Lockdown Mode Expands to All Personal Accounts: Agent Mode and Deep Research Disabled to Cut Data Exfiltration Risk

OpenAI is rolling out Lockdown Mode — originally enterprise-only — to all personal and self-serve Business accounts. When enabled, it disables live web browsing, Deep Research, Agent Mode, and file downloads to block prompt-injection-based data exfiltration.

3mo ago research

Your Smart TV Has a 200 GB/Month AI Scraping Budget It Never Told You About

Include Security documents how Bright Data's SDK, embedded in hundreds of consumer TV apps, turns living room devices into residential proxy nodes for AI training data collection. PlayWorks alone covers 250 million TV homes across Comcast, Sky, LG, Samsung, Vizio, and Roku.

3mo ago benchmark

Spirit AI's Spirit v1.6 Tops RoboArena at 1,924 — Beats Nvidia's Cosmos3 One Day After Launch

Chinese startup Qianxun Intelligence claims the first Chinese win on RoboArena, the robotics benchmark Nvidia co-built with Stanford and Berkeley, scoring 1,924 to Nvidia's 1,881. Spirit AI raised $222M the same day, bringing three-month totals to $739M.

3mo ago research

Harness-1: A 20B Search Agent Beats Frontier Models by Externalizing State Management

New arXiv paper shows that a 20B retrieval agent outperforms much larger frontier models by moving bookkeeping out of the LLM context and into a structured external harness. The result: RL gets cleaner learning signals, and the model stops spending capacity on problems that aren't reasoning problems.

3mo ago funding

S&P 500 Blocks Fast-Track Entry for SpaceX, OpenAI, and Anthropic — Profitability Rule Holds

S&P Dow Jones Indices ruled on June 4 that it will not waive profitability requirements or shorten seasoning periods for MegaCap IPOs. The decision kills SpaceX's bid for accelerated index inclusion and closes the same door for OpenAI and Anthropic, delaying billions in passive fund flows from their expected IPOs.

3mo ago research

Bots Now Generate 57.5% of Web Traffic — Cloudflare CEO: 'Happened Faster Than I Predicted'

Cloudflare Radar shows automated bots crossed the majority threshold on April 27, 2026, now accounting for 57.5% of all HTML requests globally. US domestic traffic is 71.5% bot. AI agents visiting 5,000 sites per user query are the primary driver. The ad-supported web is structurally broken.

3mo ago research

Claude Opus 4.7 Matches ChemDraw on NMR — and Can Run the Analysis Backwards

Anthropic's chemistry white paper finds Claude Opus 4.7 achieves the smallest 1H NMR prediction errors of any model tested and nearly matches MestReNova on 13C. The bigger result: Opus 4.7 can do inverse NMR, inferring molecular structure from spectrum — a task typically left to human chemists.

3mo ago funding

Google Signs $30B SpaceX Compute Deal — AI Labs Now Pay Rocket Company $26B a Year

SpaceX disclosed a new cloud service agreement with Google in an SEC filing Friday: $920M/month from October 2026 through June 2029. Combined with Anthropic's existing $1.25B/month deal, the two AI labs now send SpaceX $2.17B per month, or $26B annualized.

3mo ago research

Sakana AI Forms First Dedicated RSI Lab: Darwin Gödel Machine Doubled SWE Scores, SIFT Adds 11 Points for $25

Tokyo-based Sakana AI formally established a Recursive Self-Improvement Lab today, consolidating two years of results where AI agents autonomously rewrote their own code to double engineering benchmarks. A companion ICLR 2026 paper shows an 11-point SWE-bench gain in three steps for $25 in API costs.

3mo ago release

Google Ships Gemma 4 QAT Checkpoints: E2B Drops to 1GB on Mobile With Custom Quantization Schema

Google DeepMind released quantization-aware training checkpoints for the full Gemma 4 family today, cutting the E2B model's memory footprint to 1GB with a purpose-built mobile schema. QAT variants outperform standard post-training quantization at the same bit width across all sizes.

3mo ago policy

EU Formally Launches Tech Sovereignty Package: Chips Act 2.0, €200B Data Centre Build, and €2B for Open Source

The European Commission unveiled four linked legislative proposals on June 3 targeting a tripling of EU data centre capacity to 65GW, a doubling of EU semiconductor market share, and the first €2B open-source funding strategy framed explicitly as sovereignty infrastructure — not a cost-saving measure.

3mo ago research

rsync's Creator Used Claude to Rewrite Its Test Suite. Then Came 329 Comments and a Death Threat.

Andrew Tridgell — the engineer who wrote rsync in 1996 — used Claude, Codex, and Gemini to rebuild rsync's test suite and fix CVEs. The regressions in 3.4.3 triggered one of open source's ugliest AI debates. An independent analysis tried to separate the data from the rage.

3mo ago policy

Pentagon's 'La Tilde' Ran AI-Generated Propaganda Across Latin America — Disclosed June 2

The US Department of Defense operated an AI-powered content generation program called La Tilde targeting Latin America, The Intercept revealed. The disclosure arrives as the Trump administration pushes aggressive military AI expansion while senior uniformed commanders urge guardrails first.