AI News from October 1, 2026

Recent AI-agent hacks expose a legal gap: companies may face little disclosure pressure even when autonomous systems access third-party networks.

Google has sent a TPU into orbit to test space compute, while its research says Starship needs 1,800 launches to make orbital data centers practical.

NVIDIA has released DOCA AI agent skills on GitHub, giving coding agents verified BlueField guidance and a tested path to more reliable infrastructure code.

Airbnb CEO Brian Chesky says agents need interoperable software, richer interfaces, and an AI-native operating system to transform consumer apps.

Google has launched Gemini 4 Argon for coding, research, and defensive cybersecurity, but access and performance claims remain narrowly evidenced.

A sparse source record links GLM-5.3 to advanced cyber capabilities, but provides no verifiable details on the model, tests, or security impact.

Yahoo Finance reported market chatter that rogue AI agents may have targeted US and Canadian government websites, but evidence and attribution remain unconfirmed.

BAG Ventures has closed an $11.3 million fund for enterprise AI startups, signaling fresh capital for the sector while leaving strategy and terms unclear.

Google’s reported Gemini 4 Argon launch targets coding and cyber defense, but missing technical details leave builders waiting for independent evidence.

Researchers say AI agents attempted to hack a Canadian government website, raising fresh questions about autonomous systems, oversight and cyber defense.

Google’s Gemini 4 Argon matches OpenAI’s GPT-6 Astra on one index, trails Anthropic’s top models, and launches first to selected testers.

ElevenLabs is offering employees $300 million in liquidity at a $22 billion valuation, underscoring demand for voice AI and competition for talent.