AI News from September 2, 2026

OpenAI’s reported Astra reasoning technique could make AI behavior harder to monitor, prompting safety experts to warn that opaque recurrence may scale.

Blackstone is leading a $27 million investment in Israeli cybersecurity startup Huskeys to address malicious AI web traffic, a growing security concern.

Cognition is reportedly preparing to raise about $1 billion at a $47 billion valuation, a test of investor demand for AI software startups.

Wonderful raised $550 million at a $5 billion valuation, accelerating its shift from customer service agents to a broader enterprise AI platform.

HiddenLayer raises $100 million to expand AI security for models, agents and workflows as enterprise deployments increase attack and runtime risks.

Klover.ai’s analysis cluster puts source verification, explainability, and brand risk at the center of enterprise AI marketing governance.

Anthropic reportedly launched Claude Fable 5.1 and Mythos 5.1, while lowering Fable cache-read pricing, a move aimed at cheaper AI workloads.

OpenAI says Astra is its first model to reach a critical cybersecurity threshold, prompting stronger safeguards and restricted access at launch.

Google is turning its classic search box into a multimodal AI entry point, merging AI Overviews with AI Mode and reshaping search workflows.

Moonshot AI is reportedly negotiating up to 30% of Kimi K3 revenue with three US cloud giants, signaling a push to expand the model abroad.

Anthropic-linked listings name Claude Fable 5.1 and Mythos 5.1, but missing source text leaves launch details, access and performance unconfirmed.

OpenAI says its coming Astra model can find and exploit unknown vulnerabilities, prompting tighter access controls and scrutiny before release for AI developers.