AI Sector Daily Digest: July 28, 2026

In brief

1. Nvidia to invest $5 billion in Sutskever's Safe Superintelligence

• Nvidia will make a $5 billion equity investment in Safe Superintelligence, the AI startup co-founded by former OpenAI chief scientist Ilya Sutskever. 1
• The companies announced a strategic partnership without disclosing financial terms; Reuters says the money will expand SSI's computing resources and give it access to Nvidia's Vera Rubin hardware. 1
• The $5 billion figure came from Bloomberg and a person briefed on the deal, according to Reuters, so there is no public valuation or priced-round detail yet. 1

2. Moonshot releases Kimi K3's open weights, with a license boundary

• Moonshot publicly released Kimi K3's model details and weights after its hosted launch; The New York Times reports that people or companies using it at large scale will need licenses. 2
• Moonshot's model card describes a 2.8-trillion-parameter mixture-of-experts model with 104 billion activated parameters, native vision, and a 1-million-token context window. 3
• The weights and technical report are available through the model card, while the API remains available through Moonshot's platform; both the code and weights use the Kimi K3 License. 3

3. Microsoft launches a cyber model and an agentic security platform

• Microsoft launched MAI-Cyber-1-Flash inside MDASH, its multi-agent harness for finding and remediating software vulnerabilities. 4
• Microsoft says the combined system scored 96% on CyberGym, 12 points above Mythos, and that the compact model can handle up to 90% of tasks before harder cases are routed to GPT-5.4. 4
• The accompanying Perception platform uses agent teams for attack simulation, bug detection, triage, and remediation; TechCrunch reports that the tools are scheduled for preview availability on November 3. 5

4. Enigma raises $71 million to study how people control robots

• Enigma emerged from stealth with a $71 million seed round led by Index Ventures and Ribbit Capital, with participation from Sarah Guo's Conviction Partners. 6
• The startup is opening an online experiment in which people can interact with more than 100 proprietary AI robots housed in facilities in Israel and California. 6
• Enigma wants those interactions to reveal better interfaces and training data, but it has not disclosed a specific product use case; it says it is already working with companies in healthcare, logistics, and entertainment. 6

5. House bill narrows its AI preemption plan

• Reps. Jay Obernolte and Lori Trahan introduced a revised bipartisan bill with four cosponsors that would give the federal government more oversight of frontier AI models. 7
• The proposal would require large developers to publish safety reports, report incidents, and undergo independent audits; it also includes emergency shutdown authority for models judged major national-security risks. 7
• The revised text limits federal preemption to catastrophic-risk transparency, independent verification, and incident reporting, while leaving child safety, privacy, consumer protection, and state procurement to the states; committee action is hoped for in September, but the committee chair has not committed to a timetable. 7

What to watch next