Latest AI (Artificial Intelligence) News
Anthropic’s Claude Clears UK AISI Cyber Threshold
The UK’s AI Security Institute revealed Anthropic’s Claude Mythos Preview as the first model to pass the 32-step 'The Last Ones' corporate-network simulation, from reconnaissance to domain takeover, typically requiring 20 hours of human red-teaming. This marks frontier AI crossing into offensive cyber operations.
Chinese Labs Release Open-Weights Coding Models at Parity
Four Chinese labs—Z.ai (GLM-5.1), MiniMax (M2.7), Moonshot (Kimi K2.6), and DeepSeek (V4)—released models with agentic engineering capabilities matching Western frontiers at lower inference costs within 12 days. NIST’s CAISI shows V4 lagging US leaders by eight months on cross-domain benchmarks, but DeepSeek claims parity with top models on specific tests.
Profluent Partners with Lilly for $2.25B AI Bio Deal
Frontier AI for bio company Profluent announced a $2.25B partnership with Lilly for large-gene insertion therapeutics. Separately, embodied AI firm Sereact closed a $110M Series B funding round.
David Silver Raises $1.1B Seed for Ineffable Intelligence
DeepMind's David Silver secured $1.1B in seed funding—the largest in European history at $5.1B valuation—for Ineffable Intelligence, aiming to build superintelligence via self-play without human data. SpaceX offered a $60B buyout option preempting Cursor's $2B fundraise.
Google Releases Gemma 4 and Agentic AI Tools
Google announced Gemma 4, the most capable open model byte-for-byte, alongside Gemini Enterprise Agent Platform, eighth-generation TPUs, Deep Research Max, Google Vids, and Learn Mode coding tutor in Colab. These updates from April focus on the agentic era and business AI adoption at Cloud Next ‘26.
OpenAI Rolls Out Restricted GPT-5.5 Cyber Tool
OpenAI launched GPT-5.5 Cyber for cybersecurity testing, limiting initial access to 'critical cyber defenders,' mirroring Anthropic's strategies and sparking debate on access and competition. The tool aids in identifying vulnerabilities.
AWS Integrates OpenAI Models on Bedrock with Trainium
AWS added OpenAI models like Codex and managed agents from OpenAI and Anthropic to Bedrock, backed by multi-gigawatt commitments to Trainium silicon. This highlights AWS's AI infrastructure push.
Microsoft Copilot Reaches 20M Paid Users
Microsoft reported over 20 million paid Copilot users with rising engagement, countering adoption skepticism and underscoring demand for AI productivity tools.
2026 Declared Year of AI-Assisted Cyber Attacks
AI lowered attack barriers in 2025, enabling breaches like a 7M-user hack by a teen and exploits in under 44 days, with 28.3% of CVEs hit within 24 hours of disclosure. Attack frequency doubled, favoring attackers over defenders.
UK Launches £500M Sovereign AI Fund
The UK announced a £500 million sovereign AI fund, signaling AI as core national infrastructure amid global pushes like Germany's billions and Saudi Vision 2030. This modest investment contrasts tech giants' billions but marks a governance shift.
Robotics Achieves Compositional Generalization
Physical Intelligence’s π0.7 demonstrated generalization to unseen tasks; Skild AI acquired Zebra’s robotics business, integrating Fetch and Symmetry stacks.
OpenAI Acquires Hiro in Seventh 2026 Deal
OpenAI completed its seventh acquisition of 2026 with Hiro; Cohere merged with Aleph Alpha, Sierra bought Fragment, Qualcomm acquired Exostellar, and China blocked Meta’s $2B Manus deal.