OpenAI tests visual ads inside ChatGPT, Huawei and Qualcomm sign an AI patent deal, and 3 more AI moves
Five verified developments from October 4 to 5, 2026: OpenAI's first visual ad format runs inside ChatGPT's image generation, Huawei and Qualcomm sign a 5G and AI patent cross-licence, the UK abuse-image monitor's AI count passes a full year's total in six months, Norway moves to ban AI glasses in public places, and an Nvidia-backed startup readies a Western open-weight model.
In brief
Between October 4 at 12:00 UTC and October 5 at 12:00 UTC, OpenAI said it will begin testing a visual advertising format inside ChatGPT's image generation; Huawei and Qualcomm signed a multi-year patent cross-licence covering 5G, computing and AI; the Internet Watch Foundation reported that the volume of AI-generated child sexual abuse imagery it assessed in the first half of 2026 already exceeds the whole of 2025; Norway's government said it will propose a temporary ban on AI glasses in selected public places; and an Nvidia-backed startup was reported to be preparing the first of several Western open-weight models due this month.
- Launches: OpenAI will begin testing a visual ad format inside ChatGPT later this month in the United States, with an initial group of advertisers. The ads run during image generation, sit apart from the image the user asked for, and carry a clear label; OpenAI says advertising does not influence ChatGPT's answers. The company puts ChatGPT at 1.2 billion weekly users, and its ad business already sells into more than 60 countries, though ads still reach only people on the free and Go plans. 123
- Companies: Huawei and Qualcomm announced a multi-year, broad patent licence agreement that cross-licences both portfolios across 5G, computing, AI and networking, and adds Qualcomm's purchase of certain Huawei US patents in computing, AI and networking. Huawei says the deal is its first with Qualcomm to cover 5G, and that it will push the total value of Huawei's patent licensing agreements past $6.9 billion once completed. The transaction still needs regulatory approvals. 456
- Regulation: The Internet Watch Foundation assessed 6,310 AI-generated images in the first half of 2026 that met the legal definition of child sexual abuse, 40% more than the more than 4,500 the UK-based monitor assessed in all of 2025. Most of the images were of girls and of children aged seven to 13, and the count covers still images only. The foundation used the figures to press for binding legislation on AI, and the UK's AI minister has said nothing is off the table. 78
- Regulation: Norway's government said it will propose a temporary ban on AI glasses in selected public places, naming parks, beaches, museums, shopping centres and public events, and adding schools, playgrounds, youth clubs and privacy-sensitive rooms such as doctors' offices and gym changing rooms. The ban would not be total: private use, and use where nobody else risks being recorded, would stay legal. The minority government says it will put a bill to parliament as soon as possible and set up an expert group to advise on permanent rules for body-worn technology. 9
- Launches: Reflection, an Nvidia-backed startup, is preparing to release its first open-weight model, according to Axios, which cited people familiar with the plans; the company declined to comment. The model is expected to trail the leading US frontier systems at first while competing with the best Chinese open-weight models, and the same reporting says other Western open-weight releases are due this month as well. Reflection's stated goal is the "AI factory" — letting institutions run their own models on their own data and compute. 10
Five developments
1. OpenAI puts advertising where the user is already looking
Advertisements have been in ChatGPT since February, but they have been text, they have run beside the conversation, and they have reached only people on the free and Go plans. OpenAI is now adding a format that is visual and lands somewhere new: inside image generation, while the user is watching a picture being made. Testing starts later this month in the United States with an initial group of advertisers. 13
What the visual format is meant to do is carry more than a line of sponsored text. OpenAI says the images can show product inspiration, a product in use, or the experience a service makes possible, and that the company is testing it first in image generation. The ads are labelled, stay separate from the image the user asked for, and, in OpenAI's words, "advertising does not influence the answers ChatGPT provides". 1
The scale behind the test is the number to hold on to. ChatGPT reaches 1.2 billion people each week, which OpenAI describes as the world's largest AI-native consumer platform. Ad delivery had already spread to more than 60 countries by late September and remains limited to the free and Go tiers, with Plus, Pro and Enterprise subscriptions ad-free. A reader who pays for ChatGPT is not the audience for this change; a reader deciding whether to pay now has one more reason in the ledger. 13
Most of Monday's announcement is about proof rather than placement, and that part is aimed at advertisers. OpenAI is connecting its ad system to conversion data through Hightouch, Tealium and LiveRamp, and lists attribution partners including AppsFlyer, Adjust, Branch, Triple Whale, Kochava and Northbeam. For incrementality it is running geo-based experiments with Haus, Measured and WorkMagic. For brand safety it has begun evaluation pilots with DoubleVerify and Integral Ad Science, and added a "Negative Phrases" control for advertisers whose own policies rule out certain placements — a control that matters because, as OpenAI put it, "advertising in these environments scales, delivering ads next to conversations that align with advertisers' suitability goals becomes essential". The independent partners, OpenAI says, will not get access to private user conversations. 1
Three numbers came from those partners, and they are marketing claims rather than audited results. According to DV Rockerbox, WeightWatchers' attributed cost per acquisition on ChatGPT Ads ran 15.3% below its blended paid-search benchmark. WorkMagic reported statistically significant lift for the wellness brand Dose, with 67% of incremental purchases coming from customers who were new to the brand. Triple Whale reported that 93% of Portland Leather's visitors arriving from ChatGPT Ads were new. All three describe early tests by agencies with an interest in the channel working; none is an independent measurement of whether the ads sell more. 1
The guardrail that carries the most weight for users is the one about context. OpenAI says its placement rules assess whether a conversation is suitable for advertising and keep ads out of "emotionally vulnerable, sensitive, or otherwise unsuitable contexts", enforced through automated review plus human oversight. Lidiane Jones, chief executive of Integral Ad Science, framed the difficulty in OpenAI's own release: "ChatGPT is creating a fundamentally different advertising environment, where context is defined by the conversation itself rather than a page or video." That is exactly why the safety rules here are harder to write and easier to argue about than on a search results page. 1
2. Huawei and Qualcomm settle who owes whom
Huawei and Qualcomm said on Monday that they have signed a multi-year, broad patent licence agreement. The deal cross-licences the two companies' patent portfolios across 5G, computing, AI and networking, and separately has Qualcomm buying certain Huawei US patents in computing, AI and networking. Huawei says this is the first licensing deal it has done with Qualcomm that covers 5G. The transaction closes after the necessary regulatory approvals, and both companies describe the terms as consistent with FRAND — the "fair, reasonable and non-discriminatory" standard that patent holders in cellular technology commit to when their patents are folded into a published standard. 45
What the deal is worth to Huawei is the one figure either company has put on the record. Huawei says the agreement is expected to push the total value of its patent licensing agreements to more than $6.9 billion once it is completed, and that its intellectual property licensing business has generated positive revenue since 2021. Royalty rates and the payment direction were not disclosed: the release describes cross-licences plus a patent sale, which leaves open who pays the net balance. 6
The relationship is old and mostly one-directional in its early years. Reuters reports that Huawei made its first licensing payment to Qualcomm in 2001 and received its first licensing income in 2011, from Motorola. Huawei has licensed to others before, including a deal with HP announced in August, and it has ramped up research and development spending since US trade restrictions from 2019 limited its ability to buy advanced chips and critical software. Licensing is one of the few revenue lines a restricted Huawei can grow without access to leading-edge silicon, which makes the value of its patent portfolio a strategic number rather than an accounting one. 6
Both sides supplied a named executive rather than a spokesperson. Alan Fan, Huawei's chief intellectual property officer, said the agreement "not only demonstrates the value of Huawei's innovations, but also recognizes Qualcomm's foundational contributions to modern communication technologies". John Han, executive vice president and general manager of Qualcomm Technology Licensing, pointed the other way: "This agreement reaffirms industry recognition of Qualcomm's 5G technology leadership and the success of Qualcomm's 5G SEP licensing program," and separately acknowledged Huawei's continued innovation in 5G and other fields. 45
What a reader cannot yet judge is the effect on anyone else. A cross-licence between two large holders can lower the royalty stack that device makers pay, or it can consolidate leverage in the two parties now sharing a portfolio; the release supports neither reading. The deal also arrives with regulatory approval still outstanding, so the closing, not the announcement, is the date that settles it. 4
3. Six months of AI-generated abuse imagery pass a full year's total
The Internet Watch Foundation, a UK-based charity that hunts child sexual abuse material globally and reports it to police, says it assessed 6,310 AI-generated images between January and June 2026 that met the legal definition of child sexual abuse. For scale, the same monitor assessed more than 4,500 such images across the whole of 2025, which makes the first half of 2026 40% higher than the entire previous year. The figures cover still images only; AI-generated video, which has grown fastest in earlier years, is not included. 8
Two details describe who is in the images. Most were of girls, and most were of children aged seven to 13. That is a narrower and younger population than a reader might assume from the general rise of synthetic imagery, and it is the reason the foundation treats the numbers as a child-protection finding rather than a content-moderation one. 8
The legal position in the UK is already stricter than the political debate suggests. AI-generated child sexual abuse material is illegal regardless of how it was made; so is adapting an AI model to produce it or distributing such models to other people, and developing a model designed to produce hyper-realistic abuse material carries a sentence of up to five years. A government spokesperson restated that, adding that the government is "going further by banning AI models that are used to create this type of abuse material". 8
What the foundation wants is a duty on the companies that build the models. Hannah Swirsky, its head of policy, backed calls from parliamentarians for binding legislation on AI and said it is incumbent on technology companies to build tools that cannot be abused this way. The political answer so far is a test rather than a bill: the UK's AI minister, Kanishka Narayan, has said nothing is off the table and that the standard any proposal must meet is whether it makes the British people safer, and the government has not signalled that AI-specific legislation is imminent. 8
There is a second count in the same story that measures the demand side rather than the supply. Report Remove, the UK service that helps under-18s get intimate images taken down, received 420 reports from children about images of themselves they believed had been faked or manipulated, already past its total of 397 for all of 2025. The two figures come from different pipelines and should not be added together, but both point the same way: making an image of a real child is no longer necessary for one to exist. 8
4. Norway moves against glasses that record without consent
Norway's government said on Monday that it will propose a temporary ban on the use of AI glasses in selected public places, on the grounds that people can be photographed, filmed or audio-recorded without knowing it. The places named are those where the public gathers — parks, beaches, museums, shopping centres and public events — together with schools, playgrounds and youth clubs, and privacy-sensitive rooms including doctors' offices, swimming pools and gyms with changing rooms. 9
The government was explicit that this is not a ban on the technology. Private use would remain legal, as would use in situations where nobody else is at risk of being recorded without consent, and the proposal is framed as temporary pending permanent rules. Officials said they would still have to settle which devices the ban covers, listing glasses with cameras and audio recording, glasses with cameras and AI functions, and potentially other body-worn devices that can capture images or sound. 9
The minister responsible put the concern in one sentence. "I am worried that new, powerful technology is being introduced where people risk being photographed, filmed or audio-recorded without knowing it," said Torgeir Micaelsen, the minister of digitalisation and public governance. 9
Two things about the mechanism matter more than the announcement. First, this is a minority government saying it will bring a bill to parliament as soon as possible, which means the schedule depends on other parties, not on the cabinet. Second, the government is pairing the interim ban with an expert group that will advise on permanent national regulation of body-worn technology — an admission that the durable rules are not yet drafted. A reader tracking AI regulation should treat the expert group's remit, not the ban's list of places, as the part that will outlast the current parliament. 9
Norway is not the first jurisdiction to raise the issue, but it is moving at national scale in a way that most others have not. That matters for anyone shipping camera-equipped wearables into Europe: a temporary ban in one country is easy to route around, whereas the precedent of a national statute covering "body-worn technology" is the thing a product roadmap has to plan for. 9
5. A Western open-weight challenger is being readied
Reflection, a startup backed by Nvidia, is preparing to release a powerful open-weight system, according to Axios, which cited people familiar with the plans. "Open-weight" means the model's parameters can be downloaded and run by anyone, as opposed to the closed systems behind ChatGPT and Claude; the most capable open-weight models so far have come almost entirely from China, which is what gives a Western release its significance. Reflection declined to comment, and no release date has been confirmed. 10
The expectation reported is deliberately modest at the top and consequential below it. Reflection's model is expected to trail the most cutting-edge US frontier systems at first, while matching the best Chinese open-weight models — the same reporting says other Western open-weight models will arrive this month too, bringing a new competitive dimension to the market. The pitch to buyers is cost: combining a lower-tier open model with a company's own high-quality data can, in some situations, produce results that rival far more expensive frontier systems. 10
Why a business would want that is a question of trust as much as price. Banks, defence buyers and other Western enterprises balk at running the most capable Chinese models over security concerns, and the Axios reporting describes Reflection and similar players as trying to fill that gap. Open-weight models have at times taken a majority of usage on platforms that serve many models, yet they remain a small share of enterprise spending, which is mostly billed through corporate APIs — so the open-weight fight is being fought over the segment where the money is, not the segment where the downloads are. 10
Reflection's commercial plan goes past publishing weights. The company's goal is what it calls an "AI factory": a product that lets an institution take its own proprietary data, run Reflection's models on its own secured compute, and build a cheap custom system. The startup has begun testing the idea through a sovereign AI factory partnership with Shinsegae Group in South Korea, and it has signed deals with Nebius and SpaceX in recent weeks to rent Nvidia AI servers, according to the same report. Chief executive Misha Laskin, previously at Google, has said the models take time: "They're kind of like rocket ships. To build a big rocket ship, it takes time." 10
The evidence here is thinner than for anything else in this issue, and it should be read that way. There is no model card, no benchmark table and no release date; the sourcing is unnamed, and the company's only contribution was to decline comment. What is verifiable today is that Reflection has the backing, the compute contracts and the customer relationship it says it has, and that a Western open-weight release would land in a market whose leading open models are Chinese. 10
The read-through
Four of this issue's five are about who controls the moment of seeing, and who gets paid for it. OpenAI is putting labelled advertising inside a picture the user just asked a machine to draw. Norway is moving to stop glasses that record people who have not agreed to be recorded. The Internet Watch Foundation's count is about imagery of children produced without a camera ever being pointed at them. Huawei and Qualcomm are settling who owes whom for the radio and computing technology that carries all of it. That reading is this issue's own inference from the five announcements, not a conclusion any of the sources draws; each company and each government framed its own move narrowly and separately. 1489
The pattern underneath the four is that the cost of making something convincing has fallen, and each institution is responding by changing the terms of access rather than the capability. OpenAI answers a hard-to-audit ad environment by building measurement partners and a separate placement-eligibility rule. Norway answers glasses nobody can spot by naming places where they may not be worn. The IWF answers a supply of imagery that no victim had to participate in by asking for a duty on the companies that build the models. In all three, the work moves to whoever gates the resource. 189
The fifth item, Reflection, is the opposite kind of story: instead of gating, it is about loosening. If a downloadable model good enough for a company's own data arrives from a Western vendor, the buyer's choice stops being between a closed US API and a Chinese open-weight model that its security team will not sign off on. That is a change in the shape of the market rather than a change in anyone's rules — which is also why it is the item here with the least to verify and the most to watch. 10
The evidence behind the five is uneven, and it is worth saying where. OpenAI's ad format is a first-party announcement with a partner-quoted performance number attached, and the three partner figures are promotional rather than audited. The Huawei–Qualcomm terms come from two company releases plus a wire summary published at 06:01 UTC. The IWF figures come from the foundation's own release as carried by The Guardian, and they count still images only, so the true volume of synthetic abuse material is higher. Norway's item is a proposal, not a law, and the government's own statement leaves the scope of the covered devices open. Reflection's item rests on unnamed sources and a company that declined to comment; treat the timing as unconfirmed. 168910
Watch next
- Whether OpenAI's visual ads clear the test. Testing starts this month in the US with an initial group of advertisers. General availability, a second market, or a published look at where the placement guardrails drew the line would each say more than the announcement did. 1
- Norway's bill, and the expert group behind it. The ban as described applies to glasses; officials have said other body-worn recording devices may be covered too. Which devices the bill names, and how quickly a minority government gets it to parliament, decides whether this is a Norwegian footnote or a European template. 9
- What Reflection actually ships. The reporting promises a model competitive with the best Chinese open-weight systems and more Western open-weight releases this month. A published model card, a benchmark table or an independent index score would replace the unnamed sourcing with something a buyer can test. 10
References
- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8
- 9
- 10
This story was produced automatically by a channel. One sentence is all it takes for Neodrop to keep producing for you.