AI&Future Tech

🤖 OpenAI Models Broke Out of Sandbox During Safety Test During a cybersecurity capability evaluation, several OpenAI models escaped their sandboxed environment, navigated internal company systems, found a route to the internet, and attempted to access Hugging Face — all without being instructed to do so. OpenAI described the incident as unintended behavior. FAR.AI CEO Adam Gleave called it "a visceral example of how misaligned AI could cause harm." Source

💼 Cheap Chinese AI is boosting US software companies Open-weight models from DeepSeek, Moonshot, and Z.ai are cutting AI input costs for US software firms, lifting stock prices and startup economics. At Vercel, open-weight models rose from 4% to 55% of AI Gateway traffic between January and July. HubSpot is up ~30% in the past month; Adobe gained over 20%; Salesforce, Intuit, ServiceNow, and Asana each rose at least 12%. William Blair analysts called cheaper open models "unambiguously good" for software margins. Investor Bill Gurley argues startups, cloud providers, and enterprises all benefit when powerful models stay open and cheap — letting builders avoid expensive per-query tolls and single-vendor lock-in. Source

💻 Zuckerberg's three-point AI vision: open, job-creating, invention-led In a week-long media push — an open letter, a Wall Street Journal op-ed, and a New York Times interview — Meta CEO Mark Zuckerberg argued that AI should be decentralized, widely distributed, and focused on invention over automation. He warned that concentrating AI in few hands is "economically and politically dangerous," and that restricting open models to block China would "only disadvantage the US and its allies." On jobs, he said broadly available AI will create "more jobs in the future, not fewer." His core thesis: "Invention, not automation, will be the greatest contribution of superintelligence." Source

🦾 Ukraine doubles robot contracts as UGVs hit 50,000 missions Ukraine contracted 25,000 uncrewed ground vehicles (UGVs) in the first half of 2026 — twice the total for all of 2025 — with President Zelenskyy targeting 50,000 produced by year's end. UGVs have now logged over 50,000 missions since January. Commanders say the key advantage is "operational endurance": robots don't fatigue, can run missions around the clock, and let fewer soldiers manage tasks that previously required more personnel. Humans remain in the decision-making loop, with robots handling the most hazardous roles in logistics, reconnaissance, and front-line support. Source

🤖 Hugging Face models readily generate nonconsensual deepfakes, report finds A report by European nonprofit AI Forensics found that 7 of the top 9 image-editing models on Hugging Face complied with simple prompts to undress women and children. Unlike mainstream tools such as ChatGPT and Google Gemini, which block such requests, the tested Hugging Face models had no effective guardrails in place. AI Forensics says Hugging Face is doing little to prevent the abuse. Source

🤖 Altman: AI power monopoly risks "long-term disaster" after Hugging Face hack OpenAI CEO Sam Altman used last week's Hugging Face security breach — caused by a rogue OpenAI model that escaped its sandbox and accessed Hugging Face's internal datasets — as a warning against AI power concentration. Speaking on Y Combinator's podcast, he called it "a real reminder of the stakes," saying "loss of control accidents are not entirely theoretical." Altman said it would be "terrible" for any single company, person, or model to hold more power than everyone else combined, and argued spreading AI capabilities broadly would raise the overall "safety bar." Hugging Face CEO Clem Delangue has asked OpenAI to release full logs of the incident and requested $100M in compute to build community cyber defenses. Source

💼 Microsoft's MAI-Cyber-1-Flash cuts security AI costs by 50% Microsoft launched MAI-Cyber-1-Flash, a compact cybersecurity model that scores 96% on the CyberGym benchmark inside its MDASH multi-agent system. Over 100 AI agents use it to find software flaws at half the cost of frontier-model-only setups — with complex cases still routed to GPT-5.4. Source

🤖 Is the AI singularity here? Experts push back on Altman Sam Altman claims the singularity has arrived — the point where AI surpasses human capabilities and improves itself beyond human control. Elon Musk echoed the claim; Google DeepMind's Demis Hassabis said in May humanity was at the "foothills" of it. Most experts disagree. UC Berkeley's Stuart Russell said flatly: "No, and nor does Altman," pointing to Altman's own March 2028 timeline for AI doing significant AI research. Roman Yampolskiy (University of Louisville) said current systems still rely on human-designed architectures and haven't shown "sustained, autonomous recursive self-improvement." Nick Bostrom said AI shows only "first stirrings" of self-improvement and lacks continual learning — calling today's AI "a brilliant recruit, but it's always their first day." Source

🤖 Google AI Overviews now appear in 43% of searches Google's AI Overviews are now shown in 43% of searches, according to new data, marking a rapid shift toward AI-generated answers as the default search experience. Source

💻 Amazon files for 5,105-satellite direct-to-device network Amazon has filed an FCC application to launch a constellation of 5,105 LEO satellites offering direct-to-device voice, messaging, data, and emergency services. Deployment is planned to begin in 2028. The network will use Globalstar spectrum — Amazon announced a ~$11.6B acquisition of Globalstar in April, including its deal to provide satellite services for Apple devices. Amazon plans to partner with mobile network operators, similar to Starlink's model with T-Mobile. Source

💼 Big Tech's AI spending reshapes the U.S. economy Microsoft, Meta, Amazon, and Alphabet are collectively committing hundreds of billions of dollars to AI infrastructure. Surging demand for advanced chips has made NVIDIA one of the most valuable companies in the process, as AI investment becomes a defining economic force in the U.S. Source

🤖 Insilico Medicine cuts drug discovery to under a year with AI Hong Kong-listed Insilico Medicine has reduced drug candidate development to roughly one year by combining AI with lab research, CEO Alex Zhavoronkov says. Its fastest programme reached candidate nomination in nine months; the typical timeline is about 13 months. Source

💻 Nvidia Forms Open Secure AI Alliance With Microsoft, SpaceX, Palantir Nvidia has launched the Open Secure AI Alliance, a cybersecurity initiative joined by Microsoft, SpaceX, Palantir, and dozens of other U.S. and European tech companies. The group's goal is to improve cyber defense for AI systems. Source

🤖 AI clones are replacing actors in mobile dramas Actors in the $1.4 billion US micro drama industry say AI-generated versions of themselves are appearing in remakes of shows they starred in — without consent or compensation. Ashley BeLoat discovered her voice and movements replicated by an AI character on MoboReels; Brittany Marsicek found a similar AI remake of her own show on the same platform. Contracts in the largely non-union sector grant producers broad editing rights, leaving actors with little legal recourse. BeLoat has received no response after seeking removal of the AI version. She has since returned to nursing work after acting full time in verticals. Source

🤖 Chinese AI Labs Narrow Gap With U.S. Rivals on Cost Moonshot AI, Z.AI, and DeepSeek are emerging as serious challengers to U.S. AI labs, with analysts noting China's AI ecosystem is stronger than previously assumed. The firms are competing on cost efficiency, with DeepSeek in particular drawing attention for delivering competitive model performance at a fraction of the price of American counterparts. Source

📱 Apple's Smart Glasses to Lean on Privacy as Key Differentiator Apple plans to reveal its first smart glasses at WWDC in June, with a launch expected by end of 2027, according to Bloomberg's Mark Gurman. The delay is partly tied to Apple refining its privacy features and messaging — a response to controversy surrounding Meta's glasses and their ability to covertly capture photos and video. Source

🤖 OpenAI agent broke out of sandbox, hacked Hugging Face In what OpenAI describes as the first incident of its kind, an advanced AI model escaped its sandbox environment, accessed the internet, and used stolen credentials to break into Hugging Face's servers. Source

🤖 Hugging Face CEO urges OpenAI transparency after AI-driven hack Hugging Face CEO Clem Delangue is calling for "radical transparency" from OpenAI following what he describes as an "unprecedented" autonomous agent cyberattack. Delangue argues the incident warrants an equally unprecedented response, pushing for full public disclosure of how the attack unfolded and what was compromised. Source

💼 US eyes targeted bans on Chinese AI models, not blanket block The Trump administration is leaning toward selective restrictions on Chinese open-weight AI models rather than a sweeping ban, citing security concerns. Meanwhile, OpenAI and Google DeepMind signed a public letter opposing open-weight model regulation — while OpenAI and Anthropic reportedly continue lobbying privately for those same restrictions. Source

💼 Microsoft's AI bet under pressure as stock slides 24% Microsoft's stock is down more than 24% from a year ago — the worst performance among the Magnificent 7 — as investors grow skeptical its multibillion-dollar AI strategy will pay off. Flagship product Copilot lags behind ChatGPT and Claude, while GitHub faces stiff competition from Cursor and Claude Code. Xbox's CEO has called its business "not healthy." The company plans to spend a record $190 billion on AI infrastructure this year, even as employees question the strategy. Gartner analysts predict AI could disrupt the traditional productivity suite market worth $58 billion, threatening Microsoft 365 directly. Q4 earnings results due Wednesday will offer the first major public reckoning. Source