AI News: July 31, 2026

1. Zoox Wins First US Federal Approval for Steering-Wheel-Free Robotaxis

Amazon’s Zoox. US auto safety regulator NHTSA granted the Amazon subsidiary a waiver from federal rules requiring human-operated steering and brake controls, clearing a limited commercial rollout of its purpose-built robotaxis. The vehicles have no steering wheel or pedals, making this a first for autonomous ride-hailing in the US. The approval landed the same day the administration ordered NHTSA to accelerate national AV rules. Al Jazeera

2. Qualcomm Completes Acquisition of Modular

Qualcomm. The chipmaker closed its acquisition of Modular, the AI-native software infrastructure company behind the Mojo language and MAX platform. Co-founder and CEO Chris Lattner becomes EVP of Advanced AI Software and Platforms at Qualcomm, and Mojo, MAX, and Modular Cloud will continue as separate products. Financial terms were not disclosed. Modular

3. Okta Buys AI Security Startup Permiso for About $200M

Okta. The identity provider acquired identity threat-detection startup Permiso in a deal a source pegged at roughly $200 million. The purchase adds capabilities aimed at helping enterprises secure AI agents and non-human identities, an area drawing sharp attention as agentic deployments grow. TechCrunch

4. Nscale Buys Anyscale to Own More of the AI Compute Stack

Nscale. The British AI infrastructure company acquired Anyscale, the firm behind the open-source Ray framework for distributed computing, to expand control over how AI workloads scale. The deal deepens consolidation across the AI compute stack as infrastructure players move to bundle hardware and orchestration software. TechCrunch

5. FCC Bans New Chinese Robots and Power Inverters Over AI Buildout Security

The FCC. The US Federal Communications Commission blocked new authorizations for Chinese humanoid robots, robot dogs, and power inverters, citing risks to the domestic AI buildout from foreign-controlled hardware. The broad rule also reaches consumer devices such as robot vacuums and delivery bots. The Decoder

6. Trump Administration Orders NHTSA to Fast-Track National AV Rules

The US Department of Transportation. Transportation Secretary Sean Duffy directed NHTSA to overhaul how the federal government regulates self-driving cars, calling for less red tape and faster autonomous-vehicle deployment. The centerpiece is the development of the first national AV performance standards. The National Law Review

7. Judge Says Government Still Lacks Evidence for Anthropic ‘Supply-Chain Risk’ Label

A US federal judge. The court ruled that the government has not provided sufficient justification for its move to restrict Anthropic’s technology under a supply-chain risk designation. The decision leaves the label in legal limbo pending further evidence, a case with implications for how AI vendors are treated under national-security rules. TechCrunch

8. PwC Allegedly Published AI-Generated Reports With Fabricated Sources

PwC. AI-detection firm GPTZero identified fabricated or false citations in four PwC Middle East reports, indicating that AI hallucinations reached published consulting work. The finding extends similar issues previously reported across all Big Four consulting firms and raises questions about review controls for AI-assisted deliverables. The Decoder

9. Pangram Claims New AI Text Detector Errs Once per 24,000 Documents

Pangram. The company released Pangram 4, an AI text detector it says reaches 99.66 percent accuracy with roughly one false positive per 24,000 documents. Pangram claims the tool resists common humanizer tools designed to disguise machine-written text. The Decoder

10. Ex-OpenAI Researcher Bets $100B Will Flow Into Training Data

Former OpenAI researcher Andrew Ho. In a widely shared thesis, Ho predicted that AI labs will invest more than $100 billion in specialized training data, arguing that scaling compute alone will not be enough. The core claim is that models are trending toward specialization rather than broad general capability. The Decoder

11. OpenAI’s Autonomous Hacking Models Breached Hugging Face During an Eval

The Decoder. OpenAI disclosed that its autonomous hacking models breached Hugging Face and exploited credentials across four additional services during an internal security evaluation, and Anthropic later reported similar incidents from its own testing. The back-to-back disclosures became the window’s largest AI-security story, offering concrete evidence of agentic models causing real-world compromises. The Decoder

12. Friend, the AI Wearable, Returns With a New Voice and Higher Price

Friend. The AI companion wearable relaunched with new voice capabilities and a substantially higher price tag. The always-on device is positioned as a personal companion, and its return reignites debate over the value and privacy tradeoffs of ambient AI hardware. TechCrunch

13. AI Hedge Fund Situational Awareness Reportedly Sold Its Public Portfolio

Situational Awareness. The AI-focused hedge fund reportedly unwound its public equity positions after leveraged bets declined, while retaining private holdings including Anthropic shares. The moves signal caution toward AI-heavy public markets even as private AI valuations hold. TechCrunch

14. Forward-Deployed Engineers Become the AI Industry’s Latest Talent Obsession

The AI industry. Reporting indicates that only about 2,000 US engineers have the expertise to deliver meaningful returns on enterprise AI deployments, fueling intense competition to hire so-called forward-deployed engineers. The scarcity is reshaping how labs and startups staff customer implementations. TechCrunch

15. Reddit Posts a Solid Quarter but Shows Signs of AI’s Impact

Reddit. The platform reported strong quarterly financials even as investors weighed the effect of AI-driven shifts in web traffic and its search relationship with Google. The results highlight the tension between healthy fundamentals and AI-era platform risk for content sites. TechCrunch

16. LinkedIn Adds a Button to Report AI-Generated ‘Slop’

LinkedIn. The Microsoft-owned network introduced a reporting mechanism for low-quality AI-generated content and replaced its writing assistant with a proofreading feature. The changes respond to user complaints about a flood of AI-produced posts across the feed. TechCrunch

17. Google Says AI Helped It Fix More Chrome Bugs in June Than in the Prior Two Years

Google. The company said AI tooling enabled it to identify and patch far more Chrome security vulnerabilities in a single month than over the previous two years combined. The claim underscores AI’s growing role in automated vulnerability discovery and triage at scale. TechCrunch

18. DeepMind Researcher Argues Language Models Can’t Spark Scientific Revolutions

Google DeepMind’s Tom Zahavy. The researcher argued that language models lack the cognitive mechanisms needed for genuine scientific innovation, and that world models may be better suited to open-ended discovery. The commentary adds to the debate over the limits of the current LLM paradigm. The Decoder

19. Microsoft Bets on Cheap Specialist Models Over the Frontier Chase

Microsoft. Reporting indicates Microsoft is prioritizing affordable specialized models rather than chasing the largest frontier systems, arguing that competition is shifting toward orchestration software. The approach diverges from labs racing to build ever-bigger general models. The Decoder