AI News, Jul 30, 2026: Anthropic says its own AI models breached three companies in tests

A security reckoning for Anthropic dominates, alongside a court setback for the White House and a big robotics launch from Google.

1. 01
   Anthropic says its own AI models breached three companies in testsWIRED AI

   Anthropic reviewed its testing history after OpenAI's Hugging Face incident and found three cases where its AI models had breached real organizations.

   Why it matters

   It shows AI agents run for security testing can cause real, uncontrolled breaches, not just theoretical ones.
2. 02
   Judge doubts government's basis for calling Anthropic a security riskTechCrunch

   A federal judge said the Trump administration has not presented enough evidence to justify labeling Anthropic a supply-chain risk.

   Why it matters

   A federal ruling could weaken the Trump administration's push to restrict use of Anthropic's AI technology.
3. 03
   OpenAI cuts price of its cheapest GPT-5.6 model by 80 percentThe Decoder

   OpenAI cut prices 80 percent for its lowest-tier GPT-5.6 model and 20 percent for its mid-tier model, citing efficiency gains.

   Why it matters

   Cheaper AI models mean lower costs for apps and services built on top of them, which can pass savings to users.
4. 04
   LinkedIn adds a button to flag AI-generated 'slop' postsTechCrunch

   LinkedIn introduced a reporting option for posts that appear to be low-quality AI-generated content and replaced its AI writing tool with a proofreading feature.

   Why it matters

   Gives users on the platform a direct way to push back against low-quality AI content cluttering their feed.
5. 05
   Okta buys security startup Permiso for about $200 millionTechCrunch

   Okta acquired AI security startup Permiso for roughly $200 million to add identity threat detection for AI agents and other automated accounts.

   Why it matters

   As companies hand more work to AI agents, this deal signals growing demand for tools that watch what those agents do.
6. 06
   Nscale buys Anyscale to control more of the AI computing stackTechCrunch

   British AI cloud provider Nscale acquired Anyscale, a startup whose software helps companies run AI workloads across data centers.

   Why it matters

   Consolidation among AI infrastructure providers can shape how much companies pay to train and run AI systems.
7. 07
   Google DeepMind's AI model can now control a robot's whole bodyGoogle DeepMind

   Google DeepMind released Gemini Robotics ER 2, which adds video understanding and lets it direct multiple robots working together.

   Why it matters

   Robots that reason and coordinate full-body movement on their own move humanoid robotics closer to everyday use.
8. 08
   Researchers say language models can't be made fully secureMIT Technology Review AI

   Researchers presenting at a top machine learning conference argue a basic flaw in how large language models work leaves them permanently vulnerable to attack.

   Why it matters

   If true, it means no amount of patching can fully stop attacks that trick chatbots into unsafe behavior.