AI News: August 31, 2026

1. U.S. Adds Foreign Drones and Robots to Restricted List as China Keeps Scale

United States. New U.S. tariffs and an expanded FCC “Covered List” now target foreign-made drones and advanced robotic systems on national-security grounds, with drone tariffs taking effect in September 2026 and component tariffs in 2027. Analysts note the barriers may redirect rather than blunt Chinese competition, since Chinese firms such as Unitree and AgiBot captured 86 percent of global humanoid robot shipments in the first half of 2026. The framing points to a fragmenting global robotics market rather than a clean U.S.-China split. Source

2. NPR and NewsGuard Test Finds Chatbots Resist Foreign Propaganda Better Than Search

NPR. NPR, working with misinformation-tracking firm NewsGuard, posed 30 questions built from false narratives spread by China, Iran, and Russia between December 2025 and July 2026 to major chatbots and the largest search engines. The chatbots, including OpenAI’s ChatGPT and Google’s Gemini, mostly pushed back on the falsehoods, while AI-generated summaries appearing atop search results performed notably worse. The results offer a rare comparative benchmark of how consumer AI systems handle coordinated state disinformation. Source

3. Glassdoor Data Shows Worker Sentiment on AI Turning Sharply Negative

Glassdoor. An analysis of Glassdoor employer reviews from June 2025 through May 2026 found positive sentiment toward AI among U.S. workers fell to 43 percent while negative comments rose to 53 percent, down from 81 percent positive in 2019. Job-loss fears drove 20 percent of negative comments, followed by forced tool adoption, degraded customer experience, and workplace surveillance concerns. Roles such as claims adjusters, writers, and customer-service reps skewed most negative, while executives and software architects stayed most positive. Source

4. Caterpillar Applies Mining Automation Playbook to Enterprise AI Rollout

Caterpillar. Caterpillar is extending decades of experience running autonomous haul trucks, drills, and dozers into a company-wide AI push spanning field service, digital twins, and legacy code modernization. The company cites 1.6 million connected assets and more than 16 petabytes of structured data to train its systems, and it is spending $100 million over five years to retrain 118,000 employees for AI and autonomy roles. It is a concrete example of an industrial incumbent operationalizing AI on top of existing automation data. Source

5. SpaceX Builds Turbine Blade Foundry to Feed AI Data-Center Power Demand

SpaceX. Elon Musk confirmed SpaceX is building a turbine blade foundry in Bastrop, Texas to speed natural-gas turbine production for AI data centers, claiming in-house blade casting could cut deployment timelines by up to 18 months. The effort responds to a power crunch in which gas-turbine makers such as GE Vernova are largely sold out through 2030 as hyperscalers build dedicated generation. The expansion has drawn scrutiny over emissions and permitting, including a Virginia study estimating turbine pollution could cause several premature deaths annually. Source

6. Study Finds Coding Agents Have No Sense of Elapsed Time

Researchers. A study titled “Your Agents Are Not Time Aware,” produced by independent researchers through the MATS program, evaluated Anthropic’s Claude Code and OpenAI’s Codex and found both misjudge how long tasks take, defaulting to roughly 90-minute estimates regardless of difficulty. Claude’s estimates were off by about 3x on average and Codex by 6 to 10x, and both rated their own output around 20 points above measured results. When given tools to read elapsed time, the agents performed nearly perfectly, suggesting the gap is architectural rather than a fundamental limitation. Source

7. Bocconi Study Says AI Lifts Grades Most on the Skills It Can Fake

Bocconi University. A randomized experiment with 1,053 freshmen at Bocconi University found students using ChatGPT scored nearly a full point higher on a 1-to-5 scale, adding about two more ideas and producing more logically coherent work. A short lesson in causal reasoning diversified ideas but did not lift grades, and combining it with AI added no further grade gains. The authors, several affiliated with OpenAI, note the study measured graded output but not whether students retained knowledge without AI, leaving real learning effects unresolved. Source