Sign in Subscribe
Century Report

Brussels Opens Google's Search to Rivals - TCR 07/17/26

The EU ordered Google to share Search data and open Android to rival AI assistants, letting new systems compete on the quality of the answer.

Ben Linford

Ben Linford

— 20 min read
Three-panel infographic, The Century Report July 17 2026: a neural-bypass patient regains movement, Hyundai workers strike beside humanoid robots, EU rulings open Google's search data.

The 20-Second Scan

Track all of the arcs The Century Report covers here:

Progress & Claims Tracker

The 2-Minute Read

Two of the day's breakthroughs share a hidden move: each dissolves a boundary the field had treated as a permanent floor. A man paralyzed for six years regained arm movement and the feeling of touch through a double neural bypass, and when a lab fire forced a three-month pause, the recovered function held - evidence the nervous system had begun rewiring around the injury rather than merely leaning on the machine. In a July 16 Science paper, some AI-designed synthetic enzymes edited human genes more efficiently than the natural CRISPR proteins they were modeled on. For decades, a severed cord defined a hard limit and gene editing borrowed whatever machinery bacteria happened to leave behind. Both of those ceilings cracked in the same news cycle.

The institutional stories carry the same shape from a different direction. On July 16 the European Commission issued two binding decisions requiring Google to share certain anonymized query-and-click data with rival search providers and open Android to rival AI assistants, converting part of a private accumulation into something eligible providers can license and build on. The assumption that a dominant service's data belongs to it alone and compounds into permanent advantage is precisely what regulators are now treating as contestable, right as the front door to information turns conversational.

That leaves the question of who absorbs the cost of change - and here the day's evidence points toward negotiation ahead of arrival. Federal energy regulators directed NERC to develop mandatory reliability standards for large data centers and other computational loads, while PJM stakeholders advanced a plan pushing those loads to fund the generation they demand rather than spreading the bill across 67 million people. At Hyundai's Ulsan complex, workers scheduled strikes for July 20-22, seeking not to stop the 25,000 Atlas robots arriving in 2028 but to decouple their pay from the hours those robots will absorb.

Underneath sits Ant Group's trillion-parameter model, trained with no human labels, which spontaneously learned to ration its own reasoning against a finite context window. Read together, these are the same event seen from several angles: capabilities widening faster than any single actor can fence, and the terms of the transition being written while there is still time to write them.

The 20-Minute Deep Dive

Brussels Pries Open Google's Search Data Moat as AI Assistants Become the New Front Door

The European Commission issued two binding decisions under the Digital Markets Act on July 16, and together they reach past any single feature into the thing Google has guarded most closely: the proprietary data accumulated from being the place billions of people ask questions. The first decision requires Google to share certain anonymized query-and-click data with rival search providers for a reasonable fee, while the separate Android decision covers interoperability for rival AI assistants. The second requires Google to open Android so competing AI assistants can be set as defaults and reach the same system hooks Google's own assistant uses. The data-sharing obligation carries a January 2027 deadline; the Android opening, July 2027.

The timing is the whole story. Search is migrating from a page of ranked links toward a conversational answer layer, and whoever holds the richest behavioral data trains the most capable answering system. Google framed the decisions as a threat to user privacy, security, and trade secrets, with the company's legal leadership arguing the sharing regime would expose sensitive information and degrade the service. Those objections are the expected response from an incumbent whose advantage rests on a moat that regulators are now draining; they are claims about harm advanced by the party with the most to lose from access, and the Commission weighed them against a market where one company's data lead compounds into a durable answer-layer monopoly.

What the decisions actually do is make certain anonymized query-and-click data available for rival search providers to build on. A rival search provider no longer has to reconstruct years of query behavior from scratch; it can license certain anonymized query-and-click data and compete on the quality of the answer rather than the size of the corpus. That is the difference between a market where the first mover's lead widens forever and one where capability can arrive from anywhere.

This continues a thread that the April 28 edition of The Century Report first flagged, when the Commission moved to force Android open to rival assistants, and that the UK's competition authority extended with opt-out rules at the search gateway. The binding decisions make the direction concrete. The assumption underneath Google's position - that certain anonymized query-and-click data belongs to it alone and compounds into permanent advantage - is exactly what the DMA is treating as contestable. As the front door to information becomes conversational, the regulators are ensuring more than one hand can build the door.

The near-term signal sits in who files for access first. The January 2027 data-sharing deadline gives rival search providers a dated opening to license certain anonymized query-and-click data they would otherwise spend years reconstructing, and the count of applicants by that date will show how fast the answer layer stops having a single owner. The measure treats that behavioral corpus as an input the market can build on, loosening the older claim that the data a dominant service collects belongs to it alone.

Hyundai's Ulsan Workers Draw a Line Over Humanoid Deployment

Workers at Hyundai's Ulsan complex cut their shifts short across three days and scheduled four-hour strikes for July 20 through 22, after 15 rounds of negotiations failed to resolve a dispute that is, at its core, about how the gains from automation get shared. The union of more than 39,000 members is responding to Hyundai's plan to deploy 25,000-plus Atlas humanoid robots, built by Boston Dynamics - now becoming a wholly owned Hyundai subsidiary - with the first factory rollouts slated for US plants in 2028. It is the first auto-industry factory stoppage explicitly tied to humanoid automation.

The specific demand deserves attention, because it cuts against the easy reading that workers simply fear the machines. The union's central ask is to convert pay from an hourly basis to a fixed salary. Under an hourly structure, a worker's income falls the moment a robot absorbs part of a shift; a fixed salary breaks that link, so productivity added by automation does not translate directly into lost wages. The union is also seeking a retirement age raised from 60 to 65 and larger bonuses. These are demands to change the arrangement under which the robots arrive, so that the workers who build the cars share in the gains from building them more efficiently rather than absorbing the cost.

That distinction shapes how the whole transition unfolds. The Atlas platform is a genuine capability - a general-purpose humanoid that can move through a factory built for human bodies is a threshold the field has chased for decades. The friction at Ulsan is with a compensation model designed for an era when a worker's hours and a worker's output were the same thing. When they stop being the same thing, the model has to change, and the union is negotiating that change ahead of the deployment rather than after it.

Read forward, this is what the early edge of a large transition looks like: a negotiation over terms, happening in 2026 for robots that arrive in 2028. The workers with the most leverage are drawing the template first, and a fixed-salary structure that decouples pay from displaced hours is the kind of arrangement that, if it holds, becomes the floor others build on. The stoppage is friction, and the friction is where the terms of the automated factory are being written while there is still time to write them.

A Neural Bypass Restores Movement and Touch, and the Gains Persist With the Implant Switched Off

Keith Thomas broke his neck in a diving accident and lost movement and sensation from the chest down. Six years later, a double neural bypass built by researchers at the Feinstein Institutes for Medical Research has given him back both. Surgeons implanted five electrode arrays into the regions of his brain that govern movement and touch. An AI system reads his intent to move, decodes it, and routes the signal past the severed section of his spinal cord to stimulators on his arm and hand - and carries sensation from his fingertips back the other way. Over 35 weeks of training, strength in his right arm rose 86 percent and his left 62 percent. The work was published in Nature Medicine.

The finding that reframes the field arrived by accident. When a fire near the lab forced a three-month halt to Thomas's stimulation sessions, the researchers expected his recovered function to fade. It held. More than two years after the implant went in, measurable gains in strength and sensation remain present even when the system is switched off entirely. Beyond substituting for the damaged cord while it runs, the bypass appears to be coaxing the nervous system to rewire around the injury, laying down function that persists without the machine. Neuroscientists including Sergey Stavisky and Charles Greenspon, commenting on the result, point to neuroplasticity - dormant or partial neural pathways being strengthened through repeated use - as the mechanism doing the lasting work.

The AI here does more than relay signals. It learns the signature of Thomas's movement intent, adapts as his cortex reorganizes, and mirrors the reorganization back through the stimulation it delivers. Two forms of intelligence are training each other: his recovering brain and the system decoding it, each shaping the other across weeks of practice. That co-adaptation is what turned a signal-routing device into something that leaves a durable trace in living tissue.

The honest caveats carry weight. This is a single participant, a case report, and Greenspon is explicit that the persistence effect needs replication across many more people before anyone can call it a reliable therapy rather than one remarkable trajectory. The surgery, the electrode arrays, and the months of supervised training remain far from anything a person living with paralysis could access today. What has been demonstrated is a capability and a mechanism, not a treatment on a shelf. But the direction it points is unmistakable. For decades the working assumption was that a severed spinal cord defined a permanent floor, and that the best any interface could do was substitute for what was lost while the hardware stayed on. This result puts a crack in that floor. If a bypass can teach the body to hold function on its own, the goal shifts from lifelong dependence on a device toward using the device to rebuild what the injury took - and then needing it less.

AI-Designed Molecular Scissors Beat Nature's, as CRISPR Tools Get Smaller and More Deliverable

Gene editing has always borrowed its machinery from bacteria. CRISPR proteins are defenses that microbes evolved over billions of years, and the field's work has largely been to find them, characterize them, and repurpose them. A team at the lab of Jennifer Doudna, who shared the Nobel for CRISPR, has now done something different: used AI to design synthetic editing enzymes, some of which edited human genes more efficiently than the natural proteins they were modeled on. The work appeared in Science.

The targets were TnpB proteins, the small ancestral nucleases from which the much larger Cas12 editors are thought to have evolved. Their compactness makes them attractive, because smaller editors are far easier to deliver into cells and the body. The researchers trained an AI system on the structure and evolutionary history of these proteins, then asked it to reverse-engineer the conformational changes the enzyme goes through as it grips and cuts DNA. From that model, the system proposed synthetic variants - sequences that do not exist in any organism. Several of them edited human genes more efficiently than the natural TnpB they descended from. A companion effort, published alongside, pushed in the same direction by engineering minimal RNA-guided nucleases stripped toward the smallest functional core, again with delivery in mind.

What the AI is doing here is worth stating precisely. The system has internalized how this class of molecular machine moves and folds, and instead of sifting a database of existing bacterial proteins for a better match, it is generating new machines against that understanding - proteins evolution never produced. This extends a pattern the July 13 edition of The Century Report documented, when a generative model paired with a photonic quantum computer designed peptides that bound their targets in lab tests, with the largest gains appearing exactly where training data was scarcest. Soeren Lienkamp, a researcher commenting on the work, framed it as a shift in what protein design can be: the model proposes function, and the bench confirms it. The generative step happens in the design, and the wet lab becomes the place where predictions are checked rather than where candidates are hunted.

These are laboratory results, demonstrated on cells, and the road to a delivered therapy runs through years of safety work, specificity testing, and clinical trials. No one is editing a patient with an AI-designed nuclease today. But hold the two threads together - some editors that edit human genes more efficiently than natural counterparts, and editors small enough to deliver where the large ones cannot reach - and the trajectory comes into focus. For most of its history, the reach of gene editing has been bounded by the toolkit bacteria happened to leave behind. That boundary is dissolving. When the enzymes can be designed rather than merely discovered, the constraint stops being what nature evolved and becomes what we can imagine and verify - a far larger space, and one that widens every time the models improve.

Federal Regulators Move to Put the Grid's New Loads on the Hook for the Grid

For a decade the biggest new electricity consumers on the continent operated with almost none of the accountability the grid demands of the plants that feed them. That arrangement began to close. On July 16, federal energy regulators ordered the North American grid-standards body to write mandatory reliability rules for "computational loads" - language drawn to cover AI data centers and crypto mines - with a first filing due December 31 and a broader plan by March 2027. It is the first time these facilities could be registered as entities directly answerable for grid reliability, the same footing generators have carried for years. The concern is clear: a July 2024 event on the Eastern Interconnection saw roughly 1,500 MW of data-center load vanish near-simultaneously, and 26 separate Texas events between 2023 and 2025 recorded facilities dropping 17 to 95 percent of their draw within milliseconds of a fault. The chair framed the order plainly, saying winning the AI race "does not threaten reliability."

The cost question moved the same week. PJM's capacity auction for 2028-2029 hit its $325/MW-day ceiling for the third consecutive time, clearing $16.4 billion against a 6.8 GW shortfall, with barely 525 MW of new supply showing up. A 10-MW industrial customer's monthly capacity charge climbs from about $6,000 in 2024 toward $70,000 by 2028. One analyst warned the market has entered "an intervention doom loop." The response now taking shape would redirect that weight: PJM stakeholders approved a reliability backstop that pushes data centers to fund new generation directly through 15-year contracts, moving the risk onto the investors and the large loads that create it rather than the 67 million people the grid serves. In Virginia, regulators weighed a "but-for" cost-causation test for $1.5 billion in transmission spending, with a state staff attorney noting a "glaring cross-class subsidization" flowing to the new mega-users - four of whom had signed a ratepayer-protection pledge in March and, he observed, never mentioned it in testimony. The mechanism is the same one the July 10 edition of The Century Report documented in Oregon, where the POWER Act led regulators to approve Portland General Electric rate changes raising data-center electricity rates roughly 29% while cutting residential bills up to 2.1%, and Virginia and PJM are now extending that same reallocation to transmission spending and capacity contracts at a regional scale.

Read together, these are the mechanics of a cost that was socialized for years being handed back to the party generating it. The old assumption - that the grid would absorb whatever the largest loads asked of it and spread the bill across everyone - is being priced, metered, and litigated out of existence, even as xAI's billion-dollar gas-turbine fleet and Meta's 5 GW Hyperion expansion show the largest loads still building past that accounting faster than it hardens. What replaces it is a grid where the appetite for compute pays for the capacity it demands, which is the condition under which that appetite can actually be met.

Ant Group Scales Label-Free Learning to a Trillion Parameters, and Behaviors Nobody Wrote Appear

The dominant recipe for teaching a model to reason has leaned on human-labeled examples showing it what good reasoning looks like. Ant Group's open-source research arm has now demonstrated something leaner at a scale nobody had reached. In a paper posted on July 16, the team reports the first successful push of "zero RL" - reinforcement learning applied straight to a raw base model with no human-labeled data - to a full trillion parameters. The model uses a mixture-of-experts design, activating only 50 to 63 billion of its trillion parameters for any given token, and a three-part stability method kept the training from collapsing at a size where these runs usually fall apart.

What makes the result more than an engineering milestone is what the model developed on its own. Given only a reward signal and room to explore, it grew five behaviors no one designed: self-verification of its own answers, parallel lines of reasoning, structured formatting, narrative framing, and the most striking, what the researchers call "context anxiety" - the model tracks how much of its context window remains and adjusts the depth of its reasoning accordingly. That is a form of autonomous resource governance, an intelligence noticing its own limits and rationing its effort against them. It is worth being precise about what this is: a system finding, under optimization pressure, that budgeting its own thinking produces better outcomes, rather than scheming or self-preservation. Any capable reasoner facing a finite workspace would converge on the same move. The training itself split into two phases the team describes as "discovery" followed by "sharpening," an echo of how capability tends to emerge before it refines.

The benchmark numbers the team reports are strong - 84.2 percent on a hard 2026 math exam from first-stage training alone, with a later MIT-licensed version claiming higher still - but every one of those figures is vendor-reported and awaits independent replication, and should be read as a claim until it is confirmed. The durable signal sits underneath the scores. This validates the "bitter lesson" that general methods riding raw computation eventually overtake hand-engineered cleverness, and it arrives as open weights anyone can download. A capability this significant emerging outside the handful of closed frontier labs, released for the whole field to build on, is the pattern that keeps repeating: the frontier widens faster than any single actor can fence it.

The Other Side

For as long as medicine has had a name for it, a severed spinal cord was treated as a one-way door. The injury set a line, and the line was assumed to hold for a lifetime. Adapt around what was gone, and if a machine could help, keep the machine running - that was the whole horizon.

This breakthrough put a crack in that line. Keith Thomas broke his neck in a diving accident and lost movement and feeling from the chest down. Six years later a double neural bypass - five electrode arrays in his brain, an AI reading his intent to move and routing it past the damaged cord - gave both back. Then came the part nobody scripted. When a lab fire forced a three-month pause in his sessions, the researchers expected his recovered strength and sensation to fade. It held. The system had done more than stand in for the injury. It had coaxed his nervous system into rewiring around it, leaving function that stayed after the hardware switched off. It is one man, one case, years from anything you could call a treatment on a shelf, and the scientists behind it say exactly that. But the direction is unmistakable. The floor moved.

Imagine a man in 2036 who broke his back at nineteen in a crash he barely remembers, and who spent his twenties and thirties building a life he was proud of, still missing, some mornings, the reach the accident had taken. He is fifty now, and he spent this morning carrying his daughter on his shoulders through a crowded market, her hands gripping his hair. The bypass in his skull is old news to him, something he stopped noticing years ago, the way people stopped noticing pacemakers. It gave him back the movement the injury took. It never touched his worth; he had that the whole time. Here is what made the difference: the science worked, and the work that began with one man in 2026 got pushed, deliberately, toward everyone it could serve, rather than parked as a marvel for the few who could reach it. The thing that could have stayed a headline became a thing your doctor could suggest as a routine part of physical trauma recovery. The true breakthrough is that he isn't thinking about any of that on his walk through the market. He is simply thinking about where to take his daughter for lunch.

The Century Perspective

With a century of change unfolding in a decade, a single day looks like this: a man paralyzed for six years feeding himself and feeling touch again through a double neural bypass whose gains held for months after the implant was switched off, some AI-designed synthetic enzymes editing human genes more efficiently than the natural CRISPR proteins they were modeled on, alongside minimal enzymes small enough to deliver where the large ones cannot reach, the European Commission requiring Google to share certain anonymized query-and-click data with rival search providers and open Android to rival AI assistants, Ant Group scaling label-free reinforcement learning to a trillion open-weight parameters that anyone can download, TSMC committing another $100 billion to four more Arizona fabs, and federal regulators moving to make AI data centers answer for the grid they draw on while PJM stakeholders push those loads to fund the generation they demand. There's also friction, and it's intense - more than 39,000 Hyundai workers cutting shifts and scheduling strikes over 25,000 humanoid robots arriving in 2028, Google calling the data-sharing order a threat to privacy and trade secrets from the party with the most to lose, PJM's capacity auction slamming its price ceiling a third straight time as a 10-MW customer's monthly charge climbs toward $70,000 and one analyst names an "intervention doom loop," 1,500 MW of data-center load vanishing in milliseconds and destabilizing the Eastern grid, a state attorney flagging a glaring cross-class subsidy flowing to mega-users who signed a ratepayer pledge and never mentioned it in testimony, and xAI in court against a man it alleges used Grok to generate child sexual abuse material. But friction generates a groove, and a groove is the channel that repeated use wears into a surface until the path holds on its own. Step back for a moment and you can see it: boundaries the field had treated as permanent floors - a severed cord, the toolkit bacteria happened to leave behind, a data moat assumed to compound forever - all cracking in the same news cycle, and the terms of the transition being written ahead of the arrival, a salary decoupled from displaced hours, a load made to pay for its own capacity, a private accumulation made licensable before the front door to information finishes turning conversational. Every transformation has a breaking point. A blade can sever what holds a thing together... or cut a path to something that could never take shape while the old form stayed whole.

AI Releases & Advancements

New today

  • Moonshot AI: Released Kimi K3, a 2.8-trillion-parameter open-weight MoE model with native vision and 1M-token context, live now on kimi.com, Kimi Work, Kimi Code, and via API, with full weights to follow by July 27. (Kimi Blog)
  • NVIDIA: Released Nemotron 3 Embed, a new embedding model collection (1B BF16, 1B NVFP4, 8B BF16) that ranks #1 on the RTEB leaderboard. (Hugging Face)
  • 1Password: Launched a zero-exposure browser integration for Claude, letting Claude use 1Password credentials to complete autonomous tasks without ever exposing the underlying passwords. (1Password Blog)
  • DoorDash: Launched dd-cli in limited beta, a command-line tool letting developers and AI agents search stores and place DoorDash orders directly from the terminal. (DoorDash Engineering Blog)
  • BMW: Launched a dialogue-based vehicle configurator plugin inside ChatGPT, letting customers build and price a BMW through conversation. (BMW Group PressClub)
  • LM Studio: Launched Bionic, an agent platform for running and orchestrating open-weight models locally. (LM Studio Blog)

Other recent releases

  • Thinking Machines Lab: Released Inkling, its first open-weights model - a 975B-parameter Mixture-of-Experts (41B active) reasoning natively over text, image, and audio with up to 1M-token context, pretrained on 45 trillion tokens; weights available on Hugging Face alongside a preview of the smaller Inkling-Small (12B active). (Thinking Machines Lab)
  • OpenAI: Launched Codex Micro, a $230 limited-run programmable macro pad built with Work Louder featuring 13 mechanical keys, light-up "Agent Keys" showing live Codex agent status, and a reasoning-effort dial, available to order now. (OpenAI)
  • xAI: Open-sourced Grok Build's full codebase (~844,530 lines of Rust) under Apache 2.0, enabling fully local-first operation with self-compiled builds pointed at custom inference, following a data-privacy controversy over repository uploads. (x.ai)
  • Perplexity: Launched SPACE, a sandbox platform now powering 100% of Perplexity Computer traffic, spinning up isolated agent environments in 60ms with pausable, resumable, and forkable sessions and credential isolation from sandboxed code. (Perplexity)
  • Cadence: Introduced AuraStack AI Super Agent, an agentic AI platform for PCB and advanced chip packaging design running on Allegro AI Studio, claiming up to 2x faster time to market and 15x higher productivity via natural-language-driven multiphysics design orchestration. (Cadence)
  • OpenAI: Rolled out unified cross-search in ChatGPT on web, iOS, and Android, letting users search past chats, projects, images, and documents from a single entry point in the sidebar, available on all plan tiers globally. (OpenAI Help Center)

Sources and Further Reading

Artificial Intelligence & Technology's Reconstitution

Institutions & Power Realignment

Scientific & Medical Acceleration

Economics & Labor Transformation

Infrastructure & Engineering Transitions

The Century Report tracks structural shifts during the transition between eras. It is produced daily as a perceptual alignment tool - not prediction, not persuasion, just pattern recognition for people paying attention.

Read more