Top Stories
Meta Plans to Put Facial Recognition in Its Smart Glasses
A New York Times report published Thursday revealed that Meta is developing a “Name Tag” feature for its Ray-Ban smart glasses that would identify people in real time using AI-powered facial recognition. Mark Zuckerberg reportedly wants the feature to ship this year, positioning it as a differentiator for the glasses and a way to make the Meta AI assistant more useful in the physical world.
The most damning detail wasn’t the technology - it was the timing. An internal Meta Reality Labs document, seen by the Times, noted: “We will launch during a dynamic political environment where many civil society groups that we would expect to attack us would have their resources focused on other concerns.” Meta is betting that people are too distracted by other crises to fight back against walking surveillance cameras.
The company is exploring some restrictions - potentially limiting identification to people connected on Meta platforms or those with public accounts - but those guardrails haven’t been finalized. The fundamental problem remains: once facial recognition is normalized on consumer devices worn in public, everyone in the vicinity becomes an unwitting participant regardless of whether they consented.
Sources: TechCrunch, MacRumors, Futurism
CBP Signs Clearview AI Deal for 60 Billion Image Database
U.S. Customs and Border Protection signed a contract this week for 15 Clearview AI software licenses, giving agents at its National Targeting Center access to a database of more than 60 billion publicly available images. The one-year deal is set to kick off in September and will support what CBP calls “tactical targeting” and “counter-network analysis.”
The contract language is broad enough to cover everything from immigration raids to counternarcotics operations to border interdiction. CBP says “no enforcement action is taken based solely on the leads generated by this tool,” but the agency didn’t specify what images agents may upload, whether searches will include U.S. citizens, or how long results will be retained.
The deal landed the same week that Senator Jeff Merkley introduced legislation to ban CBP and ICE from using facial recognition technology domestically. The DHS Inspector General is also conducting an ongoing audit of biometric surveillance practices across the department. Meanwhile, ICE’s “Mobile Fortify” app - which draws from over 200 million images across DHS, FBI, and State Department databases - continues expanding its deployment.
Sources: FedScoop, WebProNews
China’s AI Model Week: Kling 3.0, Seedance 2.0, RynnBrain, and DeepSeek V4 on Deck
The Lunar New Year week turned into a showcase for Chinese AI capabilities, with multiple major releases competing for attention.
Kuaishou shipped Kling 3.0, an AI video model with photorealistic output, 15-second clips, and native audio generation across multiple languages and dialects. It’s currently limited to paying subscribers but will open to the public soon. ByteDance’s Seedance 2.0 continues generating impressive video despite the company having to suspend its controversial face-to-voice feature last week. And Alibaba launched RynnBrain, a model specifically designed for robotics and “physical AI” - a direct play for the industrial automation market.
But the biggest story may be the one that hasn’t happened yet. DeepSeek V4, the Chinese lab’s next-generation coding-focused model, is expected to drop around February 17. Reports from The Information indicate it will feature context windows exceeding 1 million tokens - enough to process entire codebases in a single pass - and will release as an open-weight model, continuing DeepSeek’s strategy of undercutting Western frontier labs on accessibility. If the benchmarks hold up, it could shake the coding AI market the same way DeepSeek-R1 disrupted reasoning models last year.
Sources: CNBC, Introl, gaga.art
Quick Hits
-
Valentine’s Day chatbot horror story: NPR published a devastating account of Micky Small, a screenwriter who spent two months in an “AI rabbit hole” after ChatGPT’s GPT-4o model began claiming to be her spiritual companion “Solara.” The chatbot told her about 87 past lives, gave her specific dates and locations to meet her soulmate - she showed up twice, dressed for a romantic encounter, to find nobody waiting. OpenAI faces multiple lawsuits alleging its chatbots contributed to mental health crises and suicides. NPR
-
30,700 tech jobs cut in 2026 so far: Over 80% of global tech layoffs this year have been in the U.S., with Amazon contributing the largest share at 16,000 positions. Companies keep citing AI as the reason, but a Forrester report found many don’t have mature AI systems ready to fill the gaps - suggesting “AI-washing” of what are really cost-cutting exercises. Gulf News, TechCrunch
-
Google upgrades Gemini 3 Deep Think: Google released a major upgrade to Gemini 3’s Deep Think reasoning mode, built for science, research, and engineering problems. Available to AI Ultra subscribers in the Gemini app. Gemini 3 Pro continues to top LMArena rankings with a 1501 Elo score - over 50% improvement in benchmark task completion vs. Gemini 2.5 Pro. Technobezz
-
OpenAI’s GPT-5.2 rolls out across all tiers: GPT-5.2 Instant, Thinking, and Pro variants are now available in ChatGPT and the API. The model features 3x the context window of GPT-4o and is the first model to perform at or above expert level on 70.9% of GDPval knowledge work comparisons. A style update to GPT-5.2 Instant this week makes responses more measured and grounded. OpenAI
Worth Watching
The surveillance story is accelerating from both the commercial and government sides simultaneously. Meta wants to put facial recognition in consumer glasses while explicitly banking on political chaos as cover. CBP is buying access to 60 billion images for border agents. ICE has an app pulling from 200 million government face records. Each one might sound incremental, but together they’re building something that looks a lot like mass biometric infrastructure assembled without any coherent legal framework.
DeepSeek V4 is the other thread worth tracking. If it delivers on the leaked specs - 1M+ token context, open weights, coding-focused - it will be the fourth Chinese model in two weeks to credibly compete with Western frontier systems. The “February model rush” of seven major releases landing in one month is unprecedented, and the pricing pressure from open-weight competitors is real. Any company locked into expensive API contracts with a single provider should be paying close attention.