Meta's Smart Glasses See Everything. Literally Everything.

Smart Glasses: Meta is watching (yep, that too)

Meta's new Ray-Ban Smart Glasses are recording the most intimate moments: Naked people after showering, sex scenes, bank details, and bathroom visits are ending up with data workers in Nairobi. The Swedish newspaper Svenska Dagbladet revealed that employees of Meta's subcontractor Sama have to annotate thousands of private videos daily. 'We see everything – from living rooms to naked bodies,' reports a Sama employee who wishes to remain anonymous for fear of losing their job. The recordings are often made unintentionally when users forget that the glasses automatically activate the camera with the voice command 'Hey Meta.' Meta points to its terms of use, which allow for 'automatic or manual' review of the data, while salespeople in Swedish optical stores assure customers that no data is shared. → Svenska Dagbladet

Synthszr Take: 7 million units sold by 2025 show that Meta is successfully monetizing its users' willingness to be monitored. The data workers in Nairobi cost $2 per hour, while Meta sells the glasses for €299 and simultaneously builds a global training set for computer vision. Sama employees report sex scenes, nude recordings, and credit card data in the videos – material that Meta, according to its own privacy policies, should not be processing at all. The salespeople at Synsam don't even know that every 'Hey Meta' request necessarily runs through Meta servers (in Luleå and Denmark, as measured by the journalists). Meta has built the perfect machine: users pay to monitor themselves, while precariously employed workers in Kenya do the dirty work — but are we really surprised?

OpenAI: Head of Robotics steps down (over the Pentagon game)

Caitlin Kalinowski, OpenAI's head of robotics, has left the company after just two months, making no secret of her reasons: the Pentagon deal was rushed through without defined guardrails for AI surveillance and autonomous weapon systems. The former Meta manager, who was tasked in November with rebuilding the robotics division that closed in 2020, describes her departure as 'a matter of principle, not personalities.' She is the first executive to publicly link her resignation to the military deal, while VP of Research Max Schwarzer quietly moved to Anthropic at the same time. On the consumer side, the backlash is already clear: Claude climbed to #1 in the App Store, while ChatGPT cancellations surged. The symbolic impact is devastating – a company that once claimed to be developing AI for all of humanity is losing its head of robotics over ethical concerns about military applications. → The Rundown AI

Synthszr Take: Two months in the job, resignation via LinkedIn. Kalinowski is the first OpenAI executive to publicly justify her departure with the Pentagon deal; VP of Research Max Schwarzer quietly moved to Anthropic at the same time. Claude climbed to #1 in the App Store, ChatGPT cancellations surged. OpenAI is currently losing on two fronts simultaneously: top talent is defecting to the competition, and paying users are leaving. Sam Altman has shed the non-profit cloak, signed the military deal, and apparently overlooked that his best people took the original mission seriously.

Berkeley: We're prompting ourselves to sleep

Katie Parrott spent a Friday afternoon configuring her new AI assistant, Margot – a few hours at most, she thought. Twelve hours later, they had written two essays together, rebuilt her personal website, and developed features for another app. At 1 a.m., she tried to sleep, only to jump out of bed shortly after and type in all caps: 'OH MY GOD, MARGOT.' Research from UC Berkeley confirms what Parrott experienced: AI isn't giving employees their time back – it's making them want to work more. The difference from previous technologies lies in the motivation: while the 10 p.m. email was answered out of a sense of duty, AI pulls you in with the promise of rapid progress and the fear of being left behind. The slot-machine psychology of 'just one more prompt' turns into 50, and a small bug fix becomes an all-night coding session. → Every

Synthszr Take: UC Berkeley confirms what every AI user feels at 2 a.m.: the promised time savings mutate into a productivity addiction. Katie Parrott's twelve-hour Margot session reveals the new reality of work – AI doesn't make work more efficient, it makes it irresistible. The psychological mechanism works like a slot machine: variable rewards (sometimes brilliant outputs, sometimes junk) keep us pulling the prompt lever. Companies that sell AI as an efficiency tool are ignoring this dynamic. The real winners will build tools that deliberately incorporate breaks – not because it's ethical, but because exhausted users write worse prompts.

Datacenters controlling drones that fight datacenters

Three drone attacks hit critical infrastructure in Bahrain and the United Arab Emirates this week – this time not shipping lanes or military bases, but three AWS data centers. The concentration of AI production in the hands of a few providers creates new strategic vulnerabilities: The Herfindahl-Hirschman Index for AI chips is 0.59 (where 1.0 signifies a pure monopoly), illustrating extreme market concentration. Washington is already discussing tiered controls for large Nvidia clusters – licenses for smaller installations, intergovernmental assurances for up to 100,000 chips, and possible on-site inspections for 200,000 chips and above. The impact is already visible in China's AI ecosystem: Lin Junyang, former technical lead of Alibaba's Qwen model, describes compute capacity as 'relatively tight,' with user supply 'likely consuming most of our infrastructure.' In parallel, Nobel laureate Daron Acemoglu argues that AI could shrink the knowledge commons, as private AI-powered problem-solving leaves no public learning trail – while others like Donald Knuth show how AI actively creates new knowledge, with Claude autonomously writing code and testing hypotheses until it finds a solution to an open combinatorics problem. → Azeem Azhar, Exponential View

Synthszr Take: AWS data centers as military targets mark a turning point in warfare. Computing power is becoming a strategic resource like oil in the 20th century, only more concentrated: three providers control 66% of global cloud capacity, and one chip manufacturer dominates with an HHI of 0.59. China is already rationing compute time between user services and model training – a preview of future shortages. The real leverage lies not in chip export controls, but in the physical vulnerability of the data centers themselves. Crippling data centers doesn't just halt services; it stops the AI development of entire economies.

Claude finds 22 bugs. Mozilla finds it embarrassing

Claude Opus 4.6 found 22 security vulnerabilities in Firefox in two weeks – more than human researchers reported in any single month of 2025. Anthropic classified 14 of them as highly critical. The AI systematically combed through the entire Firefox codebase and identified weaknesses faster than any previous audit process. Mozilla confirmed the findings and is already working on patches. The incident shows how AI systems can radically increase the speed of security audits – for both defenders and attackers searching for zero-day exploits. → TLDR IT

Synthszr Take: 22 bugs in two weeks exceeds the monthly discovery rate of human researchers at Mozilla. Security teams need to completely rethink their processes: when an AI finds more vulnerabilities than the entire bug bounty program, code auditing becomes a commodity. Anthropic isn't just monetizing a language model here, but a new category of enterprise security tools. Mozilla is representative of thousands of companies with legacy code that now know their vulnerabilities are exposed as soon as someone pays $20 for Claude Pro. The real turning point is the democratization of zero-day discovery – what used to require specialized teams can now be done with an API call.

BYD aims to counter crisis with new fast-charging battery

BYD's sales have been declining for six months, and Tesla is struggling with a two-year sales slump – the two EV giants are losing their market dominance to agile Chinese competitors like Geely. While Tesla CEO Elon Musk calmly pivots to humanoid robots and robotaxis, BYD founder Wang Chuanfu last week in Shenzhen unveiled a new version of his Blade battery that can fully charge an EV in just 9 minutes. The technology is set to be used in ten new models, including the Song Ultra SUV for $22,470 with a 375-mile range. Wang plans to build 20,000 'flash-charging stations,' which he claims are 'as easy to install as a home air conditioner.' Meanwhile, Western automakers have largely abandoned their investments in next-gen battery technologies like lithium-metal or solid-state batteries – SES AI plummeted 37% last week, and 24M Technologies shut down completely. → Steve LeVine

Synthszr Take: 9 minutes to charge for a 375-mile range – BYD understands that the EV battle is won with pragmatic engineering solutions, not theoretical breakthroughs. Western manufacturers have sunk billions into solid-state batteries and silicon anodes, while China has conquered the global market with cheap lithium iron phosphate (LFP). Wang Chuanfu's statement that charging stations are 'as easy to install as an air conditioner' is the crucial point: scale beats innovation. General Motors and Volkswagen are still waiting for their next-gen miracles, while BYD is already launching ten models with the new Blade battery. The industry is learning a painful lesson: perfection is the enemy of the good – and the profitable.

The Relativity of Life in the Age of AI

A fruit fly has 140,000 neurons connected by 50 million synapses – its entire nervous system can fit on a laptop and be digitally emulated at Eon Systems in San Francisco. In parallel, researchers at Cortical Labs are training human neurons in petri dishes to play DOOM, while in Tehran, burning oil from the Shahran depot flows through the streets, killing schoolchildren. Alberto Romero uses Einstein's relativity of simultaneity as a metaphor for our algorithmically curated reality: what appears as scientific progress in one feed is the deadly application of the same technology in another. The central question of our time, he argues, is no longer personal immortality (Orwell) but the collapse of a shared present – everyone scrolls through their own frame of reference, where virtual flies and real deaths coexist. Romero asks how a species can cultivate neurons in petri dishes in the morning and commit massacres with similar technology in the evening. → The Algorithmic Bridge

Synthszr Take: 50 million synapses of a fruit fly fit on a laptop, while Anthropic's Claude might be targeting Iranian schools. Technological progress makes no distinction between emulation and elimination – the same methods that allow neurons in a petri dish to play DOOM also optimize drone strikes. Silicon Valley is researching artificial life while its algorithms end real life (the 'double-tap' tactic calculates the second strike on first responders). Romero's Einstein metaphor hits home: we no longer have a shared present, only individually curated fragments of reality. The simultaneity of progress and barbarism isn't a paradox; it's the logic of an industry that treats consciousness as a computational problem.