Tech Digest – May 13, 2026
AI Platforms Enter the Professions
Anthropic Ships Claude for Legal — 20 Connectors, 12 Practice-Area Plugins, and a Justice Access Partnership
Anthropic launched Claude for the legal industry, packaging over 20 MCP connectors that integrate Claude directly into the software legal teams already use — document management, case research, contract review — alongside 12 practice-area plugins covering everything from litigation to regulatory compliance. The release also includes partnerships with the Free Law Project and the Justice Technology Association, explicitly targeting the access-to-justice gap by putting AI-augmented counsel within reach of people who currently cannot afford legal representation.
Note: This isn’t a chatbot for lawyers — it’s a product designed to slot into existing legal workflows through native tool integration. For any public institution running a legal department on standard practice software, the question is no longer whether AI can do legal work but whether your current tools already have an AI layer you haven’t switched on.
Sources: Anthropic
Google Fuses AI Into the Operating System — Widgets You Describe, a Pointer That Understands
Google announced Gemini Intelligence at I/O, embedding AI directly into Android’s interaction layer. Users can now describe a widget in natural language and have the system build it on-device — what Google calls vibe-coding at the OS level. Alongside this, DeepMind unveiled an AI-powered mouse pointer that understands the content it hovers over, turning the cursor from a selection tool into a context-aware interface. The prompt is no longer a text box. It’s a gesture.
Note: When AI moves from an app you open to an OS layer you point through, the digital literacy assumptions behind every citizen-facing service change. Institutions designing public portals and digital services around forms and menus are building for an interaction model that Google just made feel dated.
Institutional Trust and Workforce Repricing
Princeton Ends 133 Years of Unproctored Exams — AI Made Cheating Invisible
Princeton’s faculty voted near-unanimously on Monday to require proctoring for all in-person exams starting July 1, dismantling an honor system that had run uninterrupted since 1893. The trigger: AI has made cheating both easier to commit and harder to detect. A 2025 senior survey found that 30% of students admitted to cheating during their time at Princeton, while fewer than 1% reported a peer — the trust infrastructure the honor code depended on had already collapsed before the vote formalized it.
Note: Princeton’s honor code survived two world wars, the internet, and smartphones. It didn’t survive AI. Any institution that relies on trust-based verification — professional certifications, compliance attestations, self-reported assessments — should treat this as a leading indicator, not a university story.
Sources: Wall Street Journal, The Daily Princetonian
Hollywood Screenwriters Call AI Training Gigs “The New Waiting Tables”
Struggling screenwriters are signing on with platforms like Mercor to train AI models on dialogue, story structure, and scene pacing — the same skills these models will use to generate scripts without them. Wired reports that the work has become common enough to earn its own industry shorthand: AI training is “the new waiting tables,” the survival job that keeps creative workers afloat while the industry they trained for contracts around them.
Sources: Wired
Anthropic Approaches $1 Trillion — and Warns Investors Away from Its Own Secondary Market
Anthropic is reportedly in talks to raise up to $50 billion at a valuation of $950 billion, which would make it the most valuable private AI company, surpassing OpenAI. The round, potentially the last before an IPO, follows a February raise at $380 billion — a 2.5x valuation jump in three months. Simultaneously, Anthropic issued a public warning to investors to avoid eight unauthorized secondary marketplaces trading its shares, a sign of how much speculative capital is chasing exposure to the company.
Note: A company approaching a trillion-dollar valuation while still private, and growing fast enough that its own secondary market has gone rogue — this is the capital structure of a company that has become systemically important before it has gone public. For institutions evaluating AI vendors, the stability and governance of your provider’s capital base is becoming a procurement criterion, not just a curiosity.
Sources: New York Times, Bloomberg
Capability Threshold
GPT-5.5 Scores IQ 136 on a New Meta-Eval — and Rebuilds Programs from Scratch
Two new evaluations frame the current capability frontier in terms non-specialists can parse. The AI IQ meta-eval, which maps a calibrated mix of 12 existing benchmarks onto an implied IQ scale, crowned GPT-5.5 at 136 — well into Mensa territory. Separately, ProgramBench, an eval that tests whether models can rebuild entire software programs from scratch, recorded its first solved task: GPT-5.5 high and xhigh both completed it, choosing C and Python respectively, with xhigh dominating the broader benchmark.
Note: An IQ score is reductive, but it’s legible. A system that would qualify for Mensa and can independently reconstruct software is not a tool that waits for instructions — it’s a capability that reshapes what “junior” means in any knowledge-work context. The hiring plans and outsourcing contracts written in January were costed against a different capability baseline.
Sources: AI IQ, Kilian Lieret (ProgramBench)
Scientific Discovery and Health
Columbia Solves the Cocktail Party Problem — Brain-Controlled Hearing in Real Time
Columbia University researchers demonstrated the first real-time brain-controlled hearing system, published in Nature Neuroscience. The system reads high-resolution intracranial EEG to identify which voice a listener is focusing on in a noisy room and automatically amplifies it while suppressing the others. The “cocktail party problem” — isolating a single speaker in a crowd — has defeated conventional hearing aids for decades. This is the first system to solve it by reading attention directly from the brain.
Sources: Nature Neuroscience
Isomorphic Labs Raises $2.1 Billion to Push AI Drug Discovery Toward the Clinic
Isomorphic Labs, the Google DeepMind spin-off focused on AI-driven drug design, closed a $2.1 billion Series B led by Thrive Capital. The round includes Alphabet, Temasek, MGX, and the UK Sovereign AI Fund. The company plans to begin its first clinical trials by the end of 2026, using its AI drug design engine to compress the discovery-to-candidate pipeline from years to months. Seven pharmaceutical partnerships are already active.
Note: The UK Sovereign AI Fund participating in a drug discovery round signals that governments are starting to treat AI-driven pharmaceutical research as strategic infrastructure, not just private-sector R&D. For EU health systems watching drug pricing and supply timelines, the pipeline compression this funding enables could reshape procurement forecasts within this decade.
Sources: Reuters, Isomorphic Labs
Energy and Compute Demand
xAI Adds 19 Gas Turbines to Colossus 2 in Two Months — Bypassing the Grid Entirely
xAI has installed 19 additional gas turbines at its second data centre campus, Colossus 2, in Southaven, Mississippi over just the past two months, despite an ongoing lawsuit over environmental permits. The turbines allow xAI to generate its own power on-site rather than wait in the grid interconnection queue — a brute-force solution to the energy bottleneck that every major AI lab faces but few have addressed this aggressively.
Note: When a company installs its own power plant to avoid a queue, the queue is the problem. Grid interconnection timelines — already stretching to four years for transformers, as yesterday’s digest noted — are now so slow that the largest AI infrastructure projects are simply routing around them. For any municipality or energy regulator, the question is whether your grid planning accounts for tenants who bring their own megawatts.
Sources: Wired
Infrastructure Moves to Orbit
Google and SpaceX Explore Orbital Data Centres — While Star Catcher Builds the Power Grid to Run Them
Google is in talks with SpaceX to explore launching data centres into orbit, according to the Wall Street Journal — a move that would extend cloud infrastructure beyond terrestrial power and cooling constraints. The initiative comes as SpaceX approaches a symbolic threshold, sitting roughly 200 satellites away from having launched more than the rest of the world combined, despite giving everyone else a 61-year head start. Meanwhile, Star Catcher closed an oversubscribed $65 million Series A to build the first optical power grid in space, using laser beaming to deliver 2 to 10 times more energy to client satellites’ existing solar panels. The company already has seven power purchase agreements and a qualified pipeline exceeding $3 billion in projected annual recurring revenue.
Note: Orbital data centres need orbital power. Google exploring space-based compute and Star Catcher raising capital to beam energy to satellites in the same week isn’t coincidence — it’s supply and demand for a new infrastructure tier. Data sovereignty conversations that currently assume terrestrial jurisdiction will need to account for compute that orbits above it.
Sources: Wall Street Journal, Star Catcher
Today’s items converge on a single dynamic: AI is no longer arriving — it’s reorganizing the institutions it touches. Anthropic packages 20 connectors and legal departments gain an AI colleague they didn’t recruit. Google embeds intelligence into the pointer and the operating system becomes the prompt. Princeton dismantles 133 years of trust-based assessment because the tools students carry are smarter than the safeguards designed to contain them. Meanwhile, the physical infrastructure beneath all of this is being rebuilt in real time — xAI installs its own power plant to skip the grid, Google explores putting data centres in orbit, and a startup raises $65 million to beam power to satellites that don’t exist yet. The pattern isn’t disruption. It’s that every layer of institutional life — legal, educational, energetic, orbital — is being redesigned around a capability that most governance frameworks haven’t finished writing policy for.
