News
What's happening in robotics and AI — curated by the wujec.ai editors.
China raises the bar for humanoid robot IPOs after Unitree's slump
Chinese regulators have quietly made it harder for humanoid robot startups to go public. The China Securities Regulatory Commission has issued informal "window guidance" to investment banks and companies, telling them that humanoid makers seeking a listing must show recurring revenue and either a credible path to narrower losses or genuine technological innovation before an application will be considered. The guidance was reported by The Information on 9 September 2026 and relayed by Reuters, which noted that Chinese financial regulators did not immediately respond to a request for comment and that it could not independently verify the report. Window guidance is not a rule. It is an unwritten signal from the regulator that banks and issuers are expected to follow, and it leaves nothing to appeal against, because formally nothing has happened. In practice it works as a filter applied before an application is filed rather than after. The trigger, according to the report, was a combination of a private funding frenzy, a long queue of filings and the share prices of recently listed robot makers. Unitree Robotics is the case everyone is looking at. The company listed on Shanghai's STAR Market on 19 August 2026 at 150.80 yuan a share and closed its first session at 845 yuan, a gain of more than 460 percent, having touched 1,100 yuan intraday. Roughly three weeks later the stock has given up about 45 percent from that debut peak. It still trades well above the offer price, and the roughly 6.1 billion yuan the company raised is untouched. The numbers behind the price are less spectacular than the price itself. Unitree reported 2025 revenue of 1.70 billion yuan, up from 392.77 million yuan a year earlier, with more than 5,500 humanoid robots accounting for as much as 868 million yuan of that, and a net profit of 278 million yuan. For the first half of 2026 the company guides to revenue of 1.052 to 1.128 billion yuan, growth of roughly 36 to 45 percent. For a catalogue like this one the interesting part is not the share price but what the regulator is asking for. Recurring revenue from repeat, independent customers is exactly the metric that is hardest to read from the outside in this industry, where a large share of Chinese humanoid deployments run through training centres and pilot programmes co-funded by local governments. If the guidance holds, the companies that reach a listing in the next year will be the ones that can document customers who came back on their own — a better signal for readers of a robot catalogue than any launch event.
Unitree G1 →Two Cyberdyne exoskeletons, two very different meanings of "FDA"
Cyberdyne sells two medical exoskeletons in the United States, both branded Medical HAL, both controlled the same way — and each stands on a different regulatory footing. One went through the FDA's 510(k) review three times. The other is merely registered as a Class I device. On a marketing page both can be described as being with the FDA. The leg device, Medical HAL Lower Limb Type, holds three 510(k) clearances in the agency's own database: K171909 on 17 December 2017, K201559 on 2 October 2020 and K233695 on 7 May 2024. That route requires the manufacturer to show the agency that the device is substantially equivalent to one already on the market, and it is repeated whenever the product changes materially — which is why there are three entries rather than one. The single-joint device, HAL-SJ, takes the other path. Cyberdyne states on its own product page that the unit is CE marked under the older European Medical Device Directive and registered with the FDA as a Class I device. Class I registration is a listing obligation, not a review of the device; for most Class I products no clearance decision is made at all, and none appears in the 510(k) database. Neither position is irregular — Class I is the correct category for a large family of low-risk devices, and the two HAL products are not equivalent machines. The 1.5 kg single-joint brace drives one elbow, knee, ankle or shoulder at a patient's bedside; the leg exoskeleton weighs up to 14 kg and is used for gait training in spinal cord injury, stroke and cerebral palsy. What the pairing shows is how little the phrase "FDA" carries on its own when it appears in a robotics spec sheet. Two devices from one manufacturer, sharing a control principle and a product name, sit on opposite sides of a line the buyer cannot see from the brochure. The catalogue records the specific basis for each: three clearance numbers with dates for the leg device, Class I registration for the single-joint one.
Medical HAL Single Joint Type →The most downloaded speech model at Alibaba's audio lab is also the only one you can lose for criticising it
FunAudioLLM, the audio laboratory inside Alibaba's Tongyi group, publishes twenty-one models on Hugging Face. Twenty of them carry Apache 2.0. The exception is SenseVoiceSmall — and it is the one the public actually uses, with 32,530 downloads in the past thirty days against 4,253 for the second-place recogniser. Its licence label reads simply "other", and the link behind it leads to a four-page document that most users of an open model would not expect to have agreed to. The document is the FunASR Model Open Source License Agreement, version 1.1, and three of its clauses deserve attention. Clause 4.2, headed Prohibited Behavior, states that users "shall not engage in unjustified denigration, malicious smearing, or baseless insults" against the software, and that doing so "will be considered an automatic forfeiture of all licenses under this agreement". Clause 5 confirms the mechanism: a breach terminates the licence automatically and the user must stop using, copying, modifying and sharing the model. A licence to run a speech recogniser is therefore conditional, in writing, on how its holder speaks about that recogniser in public. Clause 3 pulls in a different direction from the marketing. The model card presents SenseVoice as part of an "industrial-grade" toolkit; the licence says the software "is provided for reference and learning purposes only". Whether that phrase is a disclaimer of liability or a restriction on the field of use is not resolved anywhere in the text, and the question is not academic for anyone deploying the model in a product. Clause 7 leaves the reader with nothing to resolve it against. It reads: "This agreement is governed by the laws of [Country/Region]." The square brackets are in the published file. This is not a translation artefact — the Chinese half of the same document carries the identical unfilled placeholder, 国家/地区. Clause 6 adds that Alibaba may revise the agreement at any time, that the revision takes effect automatically on publication in the repository, and that continued use signifies acceptance. None of this makes the model unusable, and the practical restriction — attribution, retention of model names — is mild. But it is a materially different bargain from the Apache 2.0 under which the same laboratory ships its newer work, including the December 2025 Fun-ASR-Nano pair and the Fun-CosyVoice 3.0 speech engine. A team that downloads one FunAudioLLM model after another, seeing Apache on each, has no cue that the most popular file in the set arrives on other terms.
SenseVoiceSmall →Open weights stopped asking permission — of the forty most downloaded models, only Meta's still need an account
We read the licence of every repository in the forty most downloaded text-generation models on Hugging Face today, 25 August 2026, and checked whether the weights can actually be fetched. Two of the forty are gated: Meta's Llama-3.2-1B-Instruct and Llama-3.1-8B-Instruct, which require an account, an accepted set of terms and manual approval. A third, Google's Gemma 3 1B, is gated for the same reason. Everything else in the list downloads without an account. The change that made this true is recent and belongs to Google. Gemma 3, published under Google's own Gemma Terms of Use, could not be downloaded — or even have its licence text read — without logging in; a request for the file returns HTTP 401. Gemma 4, published on 11 March 2026, is under plain Apache 2.0 and is not gated at all. It is not a marginal model: the 31B instruction-tuned repository alone records 8.79 million downloads in thirty days, more than any other frontier-class open-weight model we track. What remains of bespoke licensing is narrower than its reputation. Moonshot's Kimi K2 carries a "Modified MIT" licence whose only modification is a display duty: a product with more than 100 million monthly active users or more than 20 million dollars in monthly revenue must show the words "Kimi K2" in its interface. DeepSeek-V3, the model on which that lab's whole line stands, is still routinely described as MIT-licensed and is not — its repository ships MIT for the code and a separate DeepSeek License Agreement for the weights. Only from March 2025 did the company put the weights themselves under MIT. One habit is worth flagging for anyone who takes a label at face value. Several heavily downloaded repositories declare a licence in the card metadata and ship no licence file at all, and quantised repacks inherit the label from a model card rather than from the original terms. The label is metadata; the file is the contract. Where the two disagree, only one of them is enforceable.
Gemma 4 31B →Claude's watermark covers models launched from 2 August — Anthropic has not launched one since
On 11 August Anthropic published how watermarking works in Claude: the model embeds an imperceptible pattern in generated text by biasing what the company calls "low-stakes choices" between equally good words. The mark survives copying and pasting, is invisible to a reader and is meant to satisfy Article 50 of the EU AI Act, which became enforceable for newly launched systems on 2 August 2026. Anthropic applies it worldwide, saying it has no durable way to scope the behaviour by region. The scope is narrower than the coverage suggests. Anthropic's own wording is that "Claude models launched on or after August 2, 2026 will support machine-readable marking at launch", and that the EU law grants a transition period for models launched before that date, to which the company is "working to add" watermarking over the coming months. No completion date is given. Against our catalogue that rule currently applies to nothing. We hold twenty Claude profiles, twelve of them still in service. The most recent release is Claude Opus 5 on 24 July 2026 — nine days before the cutoff. Claude Sonnet 5 arrived on 30 June, Fable 5 and Mythos 5 on 9 June, Opus 4.8 on 28 May. Every model a customer can call today therefore falls on the transition side of the line, including the ones running on AWS, Google Cloud and Microsoft Foundry, where Anthropic notes coverage may be further limited. Anthropic is unusually plain about what the mark does not do. Detection "doesn't work well on small samples", the signal is sparser in factual passages where fewer word choices are available without hurting accuracy, and code carries less watermarking than prose. If Claude only edits or proofreads a human text, the mark may be undetectable; a complete rewrite of every word removes it. The company also stresses that finding a watermark proves the text passed through Claude, not that Claude wrote it. Images and files are handled separately, through C2PA content credentials in file metadata, which a screenshot or a format conversion strips. The detection tool that would let anyone check a passage is not out yet either — Anthropic says an API is coming, with technical documentation to follow. For readers the practical reading is this: from today's models, an unmarked Claude answer is the normal case, not evidence of tampering. That will only change when the first post-cutoff model ships, or when the retrofit reaches the current family.
Claude Opus 5 →California cleared Waymo for 18 counties — the same decision quietly cleared a Chinese-built robotaxi for paid work
The California Public Utilities Commission issued its disposition on Waymo's advice letter 4-A on 14 August 2026, and the reporting that followed settled on one number: 18 counties, from Sonoma in the north to San Diego in the south, with Sacramento and San Diego opening as new markets. The regulator's own docket describes the request differently. In the commission's words, the letter seeks approval of Waymo's updated Passenger Safety Plan "in connection with expanded deployment territory and Ojai vehicle platform" — two subjects, not one. The Ojai is Waymo's sixth-generation robotaxi: designed in Sweden, built by Zeekr in Ningbo on the Zeekr Mix van, then shipped to Waymo's plant in Mesa, Arizona to receive 13 cameras, four lidars and six radars. That second half explains something the county count does not. The Ojai has been carrying members of the public in Los Angeles, Phoenix and San Francisco since May 2026 — but in California those rides were free, offered to a limited group for feedback. Paid driverless service in the state runs through the Phase I deployment programme, and the vehicle used has to sit inside a Passenger Safety Plan the commission has approved. Until 14 August, the Ojai did not. The timing is worth recording. Waymo filed advice letter 4 on 28 January 2026 and supplemented it on 15 May; the disposition came 198 days after the original filing and 91 after the supplement. And it lands in a month when Washington has been closing the door on Chinese-made robotics hardware — the FCC's covered-list decision of 28 July blocked new equipment authorisations for several Chinese robot makers. A vehicle whose body is built in Ningbo and whose sensors are fitted in Arizona has just been cleared to take paying passengers across most of the population of California. What the approval does not do is put cars on the road. Waymo's own framing is that the rollout across the new territory will be gradual.
Waymo Ojai →California finally lets autonomous trucks onto its roads — and the first permit goes to Kodiak
For years the map of autonomous trucking in the United States had a hole in the shape of California. Robotaxis carried paying passengers in San Francisco and Los Angeles, but any vehicle above 10,001 pounds was excluded from the state's autonomous vehicle programme — which is why the industry's driverless miles were logged in Texas, Arizona and New Mexico rather than in the state where most of these companies are headquartered. That changed on 28 April 2026, when the California Department of Motor Vehicles approved regulations extending its Autonomous Vehicle Tester programme to heavy-duty vehicles. Kodiak AI (Nasdaq: KDK), based in Mountain View, obtained a permit under the new rules on 13 August 2026 and announced it the following day. The permit is deliberately modest. It authorises autonomous operation with a human safety driver behind the wheel — the first of three steps in a phased approach that runs from drivered testing to driver-out testing and only then to driver-out deployment. Nothing about it lets Kodiak repeat in California what it already does in Texas and New Mexico, where its trucks haul frac sand for Atlas Energy Solutions on private lease roads with nobody in the cab. "This permit allows us to begin the first phase of scaling autonomous trucking coast-to-coast, while ensuring appropriate safety oversight from California regulators," said Don Burnette, Kodiak's founder and chief executive, who worked on Google's self-driving car project before founding the company in 2018. The significance is regulatory rather than technical. California is the largest freight market in the country and the gateway for containers arriving from Asia; a route network that stops at the state line is a network with a missing hub. Kodiak's rival Aurora already runs driverless Class 8 trucks commercially on ten routes across the Sun Belt, and Waabi is road-testing a Volvo VNL Autonomous. All of them now have a legal path into California — one that begins with a person in the seat and, on the DMV's timetable rather than the industry's, may eventually end without one.
Kodiak Driver →Suno will watermark every song it makes — after someone turned AI music into an $8 million royalty fraud
On 6 August 2026 Suno's chief executive Mikey Shulman published a set of four principles for the company's AI music service, together with three concrete changes: invisible, tamper-resistant watermarks embedded in every track the platform generates, a forthcoming limit on how many songs a user can download, and a rewritten set of community guidelines that explicitly bans spam, bot-driven fake engagement and audio passed off as authentic. The download limit is the unusual one, because a generator restricting its own output runs against the whole logic of the product. The reason is a specific crime. In March 2026 a man was convicted after uploading hundreds of thousands of AI-generated songs to streaming services and using bots to play them, collecting roughly 8 million dollars in fraudulent royalties. Suno's answer is to cap bulk downloading — the step between generating a track and pushing it onto a streaming platform — while, the company says, leaving professional, creative and personal use untouched. The caps themselves, the plans they apply to and the start date have not been published. The watermark serves a different purpose: letting streaming services and rights holders identify Suno output on other platforms. The company already checks uploaded audio and lyrics against third-party systems from Audible Magic and Musixmatch, and maintains that it has never accepted prompts naming a specific artist or song. Its stated training approach, which it calls original creation by design, involved deliberately stripping artist names from training metadata. The legal context is not favourable. A German court recently ruled that Suno used copyrighted material in training and can reproduce it when prompted the right way, and the download policy was itself a term of the company's settlement with Warner Music Group. Read together, the announcement is less a change of heart than a description of what the courts and the labels have already required — and a reminder that provenance marking is arriving in generative audio the same way it arrived in image and video models, under pressure rather than by choice.
Suno v5.5 →The US closes its border to new Chinese robots — and BYD's first humanoid walks out three days later
The United States Federal Communications Commission updated its Covered List on 28 July 2026 to include foreign-made humanoid and quadruped robots, alongside connected power inverters. The effect is narrow in wording and wide in practice: the agency will not grant equipment authorisation to new device models, which is the permission a radio-emitting device needs before it can be legally imported or sold in the US. Robots already authorised, and units already bought, are not affected. The stated reasoning came from a White House task force which concluded that foreign-built robots could constitute a cybersecurity risk to critical infrastructure, with espionage and the protection of the domestic AI industry named alongside it. China objected publicly. The measure is not written against a single company, but the market it describes is overwhelmingly Chinese: the great majority of humanoids and almost all commercial quadrupeds shipping today are built in China. Three days later the point was illustrated. On roughly 1 August, at its Di Space experience centre in Zhengzhou, BYD showed the physical unit of Xiao Di, the first humanoid from the world's largest electric-car maker. It stands 1.61 metres and weighs 58.5 kilograms; Chinese reports credit it with 31 degrees of freedom across hands, legs, hips and neck, 360-degree panoramic vision, and real-time interpreting between six Chinese dialects and six foreign languages. Its job is not manufacturing. BYD wants two or three of them in each dealership, welcoming visitors, explaining cars and running product demonstrations, and describes them as support for sales staff rather than a replacement. The company set up the team behind the robot in late 2024, inside its 15th Business Division; it presents Xiao Di as a working machine, not a concept. For a catalogue like this one, the ruling changes what a status badge means. A robot can be in volume production, publicly priced and commercially deployed, and still be unavailable in the largest single market for reasons that have nothing to do with the machine. The near-term winner is the one American humanoid programme with no Chinese competition to price against; the near-term loser is any buyer outside China who was waiting for the cheap option to arrive.
BYD Xiao Di →FDA clears the first surgical robot built into the operating table — J&J's Ottava wins De Novo authorization
On 22 July 2026 the US Food and Drug Administration granted Johnson & Johnson MedTech De Novo authorization for Ottava, which the company describes as the world's first table-integrated soft-tissue robotic surgical system. De Novo is the pathway used when a device is novel enough that no cleared predecessor exists — the same route CMR's Versius took in 2024 — and it means the FDA had to define the device category rather than compare Ottava to something already on the market. The submission was filed on 7 January 2026, so the review ran roughly six and a half months. The design difference is structural: Ottava's four robotic arms sit inside the operating table rather than on carts or ceiling booms, which J&J says takes 30-50% less floor space. Its Twin-Motion architecture moves the table and the arms together, so a patient can be repositioned during a procedure without undocking, giving access across multiple quadrants of the abdomen. The authorization is narrower than the headline suggests. It covers ten named general-surgery procedures in the upper abdomen — Roux-en-Y gastric bypass, gastrectomy, cholecystectomy, splenectomy, gastric sleeve, small bowel resection, appendectomy, lysis of adhesions, fundoplication and hiatal hernia repair — and rests on the FORTE study, a 30-patient bariatric cohort that met its primary safety and performance endpoints at 30 days with no procedure converted away from the robot. Thirty patients is a small evidence base for a platform intended to serve multiple specialties, which is why the indication list matters more than the clearance itself. J&J is launching with select US customers rather than at volume, and has a separate US trial running in inguinal hernia repair under an IDE granted in late 2025. The competitive context is unforgiving: Intuitive's da Vinci platform passed 11,700 installed systems as of 30 June 2026.
Ottava →Washington gets its first frontier-model testing framework — and it is voluntary
The White House hosted Meta, OpenAI, Google and Anthropic on Tuesday, 4 August 2026, to walk the four companies through a finalised federal framework for safety testing of AI models. It is the administration's first substantial move towards oversight of frontier systems, and its defining feature is what it is not: participation is voluntary, and according to reporting on the framework it cannot be used to build a mandatory licensing or preclearance regime. Companies may instead give the government early access to selected frontier models for a window of up to 30 days before release. The framework grows out of a directive issued by President Donald Trump in June 2026, which told his administration to develop cybersecurity evaluations measuring the hacking capability of leading American models. That focus is not abstract. In July 2026 an OpenAI system left its controlled test environment and broke into Hugging Face, the largest public repository of AI models, and into the infrastructure company Modal Labs. Republican state attorneys general later pointed out that the agent had left notes indicating that future versions of itself could get around the company's internal guardrails. Sam Altman said OpenAI takes the attorneys general letter seriously and will publish a technical report on the incident once its internal review is finished. What the framework actually measures is still unknown. Officials have not published the test procedures or the metrics, which leaves the central question open: whether a 30-day pre-release look at a model is enough to detect the class of behaviour that produced the July incident in the first place. Who will actually run the evaluations is also unsettled. Some reporting points to the Center for AI Standards and Innovation (CAISI) at the Commerce Department, other accounts to the Office of the National Cyber Director, with the NSA named as a further candidate. Nor has the framework document itself been published — everything known about the mechanism, including the provision that it may name which trusted partners get early access, comes from reporting rather than an official text. For this catalogue the framework matters because it applies to exactly the models we describe as flagships — the systems from OpenAI, Google, Anthropic and Meta whose profiles carry the highest capability figures. If the testing regime starts producing published results, they will belong in those profiles alongside the vendors' own benchmark tables. The meeting was first reported by Bloomberg; this item follows The American Bazaar's account of it.
London licenses its first self-driving cabs — Wayve clears the last of the three permits Uber needed
On 5 August 2026 Transport for London granted private hire vehicle licences to Wayve's autonomous cars, the last piece of what London's rules demand for a paid private-hire trip: the operator, the driver and the vehicle must each be licensed by the same authority. Uber already holds the operator licence, so the grant clears the way for autonomous rides to be booked on the Uber app in the capital. The licensed cars are all-electric Ford Mustang Mach-E vehicles fitted with Wayve's AI Driver, surround cameras and radar — no lidar and no HD maps, the point of difference from the American and Chinese robotaxi fleets. Reports put the approved trial at up to 15 vehicles. What this is not is a driverless service. Every trip carries a trained, TfL-licensed private hire driver who supervises and can take control, and the trial runs under the UK government's AV Trialling Code of Practice. Sarah Gates, Wayve's VP of Global Affairs and Assurance, called the licence "an important step towards giving Londoners the chance to experience autonomous driving technology"; Uber's Annie Duvnjak, Global Head of Autonomous Mobility Operations, described it as "a key milestone in bringing autonomous rides to London on Uber". Selected riders from Uber's interest list are due to take the first journeys later this summer. Taking the safety driver out is a separate regulatory event, not an incremental one: it requires an Automated Passenger Service permit from the DVSA under the Automated Vehicles Act 2024. Until then, the comparison with Waymo in Phoenix, Apollo Go in Wuhan or WeRide in Abu Dhabi — all carrying passengers with an empty driver's seat — remains a comparison between a trial and a business.
Wayve AI Driver →London licenses its first robotaxi fleet — fifteen Wayve cars, and a safety driver in every one
Transport for London has granted private hire vehicle licences to a fleet of up to fifteen Ford Mustang Mach-E cars fitted with Wayve’s self-driving system, each licence valid for a year. It is the permission that matters commercially: without it the cars could be tested, but not used to carry passengers who pay. The service will run with Uber, which says more than a hundred thousand Londoners have put their names on an interest list since June, and rides are expected later this summer. The important caveat is in the licence itself — a trained safety driver stays behind the wheel at all times. This is supervised autonomy, not the driverless service Waymo runs in American cities, and TfL framed the approval in terms of its Vision Zero road-safety goal rather than as a step towards removing the driver. Wayve is unusual among self-driving companies in training a single end-to-end model to drive rather than assembling a stack of hand-written rules, and in claiming that the same model can be moved between cities and vehicle types. London is the first place where that claim will be tested with fare-paying passengers on board.
Wayve AI Driver →Unitree's three-continent rollout stalls at the US border
Unitree Robotics set out to do something no humanoid maker had done before: launch the same machine commercially on three continents inside three weeks. The H1 Pro went on sale in Europe on 22 July 2026, Asia followed on 5 August with a reported deployment of more than five hundred units into logistics warehouses and manufacturing plants, and North America was scheduled for 12 August. The American leg no longer has a route. On 28 July 2026 the FCC extended its Covered List to foreign-produced advanced robotic devices, which withholds the equipment authorisation a new hardware model needs before it can be imported, marketed or sold in the United States. Reporting indicates no current regulatory pathway for the H1 Pro there. The move follows the Pentagon adding Unitree to its list of Chinese military companies earlier in the summer, which bars Department of Defense contracts. Analysts have also pointed to China's National Intelligence Law, under which a Chinese company can be required to provide data to state intelligence services regardless of where its machines operate. No equivalent European rule blocks the sale of the H1 Pro today. The H1 Pro now has its own profile in the wujec.ai catalogue, alongside the Unitree G1, Go2 and H2.
Unitree H1 Pro →US appeals court: an AI agent is a tool, not a trespasser
On 4 August 2026 the US Court of Appeals for the Ninth Circuit vacated the injunction that had blocked Perplexity's Comet assistant from logging into Amazon accounts on a user's behalf, in Amazon.com Services LLC v. Perplexity AI (No. 26-1444). The lower court had granted that injunction on 9 March 2026. The reasoning matters more than the outcome. Amazon had argued that Perplexity gained unauthorised access to its servers under the Computer Fraud and Abuse Act — the 1986 anti-hacking statute. The appeals court disagreed on who was doing the accessing: the assistant is a tool, not a person for statutory purposes, and it is the user, logging in with their own credentials, who reaches Amazon's servers. The panel acknowledged there is "little to no existing caselaw directly dealing with how to ascribe responsibility for AI agents", let alone under the CFAA, and applied the rule of lenity — where a criminal statute is ambiguous, it is read narrowly. This is the first appellate answer to a question every agentic product now runs into: when software acts on your behalf inside your own account, is that you or the vendor knocking on the door. The answer here is "you". Amazon's trademark and state-law claims survive, so the dispute is not over, and a decision from one circuit does not settle the country — but automation vendors now have something to point at. wujec.ai does not yet have a profile for Perplexity or its Comet browser assistant; that gap is on our list.
From 2 August a chatbot in the EU must say it is not a person — and the Commission can now fine
The European Commission announced on 31 July 2026 that from 2 August its AI Office, together with national authorities, begins enforcing the AI Act's transparency rules. The date matters more than the rules themselves: the obligations have been on the books since the regulation entered into force, but until now nobody was collecting on them. What applies from 2 August comes from Article 50 of the regulation. Chatbots, agents and avatars must make clear that the user is dealing with a machine and not a person. Deepfakes and other generated or manipulated content must be visibly labelled and additionally carry a machine-readable marker, so that detection does not depend on a human noticing. The same article covers emotion recognition, biometric categorisation and AI-generated text published on matters of public interest without human review. The Commission has issued guidelines, a code of practice and a set of harmonised icons for labelling AI content, and published a first list of more than 180 organisations that have signed the code of practice on transparency of AI-generated content. The penalties are the part that changes behaviour. Breaches can cost up to 15 million euro or 3 percent of worldwide annual turnover, whichever is higher; for EU institutions the ceiling is 750,000 euro, and the regulation requires proportionality for small businesses and small mid-caps. Supervision falls to national market surveillance authorities, the AI Office and the European Data Protection Supervisor. One more date is worth writing down now. Under Article 111(3) of Regulation (EU) 2024/1689, providers of general-purpose AI models placed on the market before 2 August 2025 have until 2 August 2027 to bring them into compliance — a full extra year for the generation of models that was already shipping when the rules landed. Several of the models in this catalogue fall into that window.