General AI News (Oct 5): ChatGPT Ads, DeepSeek Gap & Norway Ban
Oct 5 AI news: OpenAI starts testing ChatGPT ads inside image generation, the US model lead over China narrows to 3%, and Norway moves to ban AI glasses.
Monday's AI news is about the commercial and geopolitical plumbing of the industry. OpenAI started testing ads that appear while you generate an image, turning its most-used free surface into inventory. Bloomberg Intelligence put the US lead over China's best model at roughly 3%, the narrowest it has tracked. A $20 billion Nvidia deal drew a shareholder suit, Qualcomm agreed to pay Huawei for AI patents, and Norway moved to ban AI glasses in public places.
Yesterday's edition covered the California subpoena and Amazon's data-center pushback, and the AI news archive holds the rest of the week.
OpenAI tests ChatGPT ads inside image generation
ChatGPT is OpenAI's assistant, and on Oct 5 the company said it will start testing a visual ad format in the United States later this month, showing clearly labeled images of sponsored products and services on screen while a user generates an image. Source: OpenAI Source: The Verge
The ads sit outside the generated image, will not influence answers, and stay off Plus, Pro and Enterprise accounts. Measurement runs through Hightouch, Tealium and LiveRamp, with brand safety from DoubleVerify and Integral Ad Science, two ad-verification vendors. It is OpenAI's first ad unit built on images, after it put text ads in ChatGPT in February.
Bloomberg Intelligence: the US lead over China's best model is about 3%
DeepSeek is a Hangzhou lab that ships open-weight models. On Oct 4 Bloomberg Intelligence published a note from senior analyst Robert Lea putting the US edge over China's best model at roughly 3% on LiveBench, the smallest gap it has recorded: DeepSeek's V4.1 Flash scored 81.1 against 83.4 for Anthropic's top model, a 2.3-point spread. Source: Bloomberg
The gap was about 9% in May and 15% earlier in 2026. DeepSeek scored 77.3 on the agentic coding subtest against 66.1 for Anthropic, though only three of the top 15 LiveBench entries are Chinese. Lea credits Chinese labs tuning for domestic hardware, which weakens the case for US export controls and for Western labs' pricing power.
Ex-Groq engineers sue over Nvidia's $20 billion deal
Groq is an AI chip company whose founders moved to Nvidia under a licensing deal. The Financial Times reported Oct 5 that former Groq engineers Benjamin Serebrin and Joshua Rubin filed a shareholder suit in Delaware claiming the December 2025 Nvidia transaction short-changed common holders and left Groq a shell. Source: FT
The reported structure is about $17 billion in licensing fees plus roughly $3 billion of Nvidia stock bonuses for about 200 engineers moving over, with the complaint naming founder Jonathan Ross and president Sunny Madra. The DOJ opened an antitrust probe in September, and senators Warren, Blumenthal and Wyden have criticized licensing-shaped acquisitions.
Qualcomm will pay Huawei for AI and 5G patents
Huawei is the Chinese telecom and chip giant under US restrictions. Qualcomm announced Oct 5 a multi-year cross-license covering 5G, compute, AI and networking, plus its purchase of certain Huawei US patents. Nikkei reported it is the first time Qualcomm is a net payer to Huawei, and IPFray estimates the bought portfolio at about 681 patents, 0.23% of Huawei's 296,248 active patents and applications. Source: Qualcomm Source: Nikkei
Google freezes its open-source bug bounty over AI slop
Google's OSS VRP pays researchers for vulnerabilities in open-source projects. The Verge reported Oct 5 that Google paused the program until at least next year, citing a "significant rise" in AI-generated submissions, "the vast majority of which are not valid," that swamped triage. Linux maintainers have reported a similar flood of low-quality AI security reports, which delays fixes for real bugs. Source: The Verge
Norway moves to ban AI glasses in public places
Norway's center-left government said Oct 5 it will bring a bill to parliament for a temporary ban on camera-equipped AI glasses in parks, beaches, museums, shopping centers, schools, kindergartens, healthcare facilities, gyms and public events. Source: Reuters Source: Ars Technica
Digitalisation minister Torgeir Micaelsen said the pause "will give us time to conduct a thorough assessment," and an expert group will draft permanent rules. Oslo already banned camera glasses in schools, and Equinor banned them in offices and offshore. It is the first national crackdown on AI wearables, and it lands on Meta's Ray-Ban line.
IWF: AI child abuse imagery is up 40% in six months
The Internet Watch Foundation is a UK charity that hunts child sexual abuse material online. It said Oct 5 that analysts assessed 6,310 AI-generated images meeting the legal definition of that material between Jan 1 and Jun 30, 2026, which is 40% more than the 4,512 found in all of 2025. Source: IWF
Children aged 7 to 13 were 79% of detections, up from 70% in 2025, and 98% of images with recorded gender showed girls. The IWF says hash matching cannot catch newly generated content, and it wants the EU's delayed Child Sexual Abuse Regulation passed.
Cohere puts enterprise agents on a budget
Cohere is a Toronto enterprise AI company. Its North 2 release on Oct 5 adds cross-session memory, shared knowledge libraries, reusable skills and hard spend controls: per-user and per-team quotas, rate limits, an organization-wide cap, and automatic downgrades to cheaper tools for simple tasks. It runs in the cloud, on premises or fully air-gapped, and lets customers plug in models from Anthropic, OpenAI and Google, or cheaper open weights from Alibaba, DeepSeek and Z.ai. Source: VentureBeat
Cohere cites VB Pulse findings that 21% of enterprises monitor agent spend only after the fact, and 68% traced a confidently wrong agent answer to missing business context.
Washington state rallies against AI data centers
At least 19 protests across Washington state on Oct 4 called for a moratorium on large AI data centers ahead of the January legislative session, with rallies in Seattle, Spokane, Tacoma, Olympia and Moses Lake. Source: Seattle Times
Seattle's council passed a one-year moratorium in June, and Skagit and Snohomish counties plus Renton, Spokane, Burien, Federal Way and Marysville have temporary prohibitions. A state regulation bill passed the House in spring and stalled in the Senate amid Microsoft and Amazon lobbying. Data Center Watch counts $64 billion of US projects blocked or delayed by local opposition.
Meta says Muse Spark co-authored six math papers
Meta announced on Oct 2 that mathematicians used Muse Spark 1.1 and 1.2 in Thinking Mode through ordinary meta.ai chat, with no custom research scaffold, to co-author six papers, five of which answer previously open questions. Source: Meta Research
They include finite-time blow-up in the mass-critical biharmonic nonlinear Schrodinger equation, open since 2015, and a 384-element counterexample to Kida's 2024 conjecture found by a GAP search the model generated. Each paper labels AI-drafted and human-drafted passages, and a second team reviewed the work. Reviewer Jason Dean Lee disputes the framing, arguing only about three were genuinely open.
Quick Comparison
| Entity | Key Metric | What Changed | Source |
|---|---|---|---|
| OpenAI | First image ad unit | Visual ads test in the US this month | The Verge |
| DeepSeek | 2.3-point LiveBench gap | US lead over China's best model at 3% | Bloomberg |
| Nvidia and Groq | $17B licensing plus $3B stock | Ex-engineers sue in Delaware | FT |
| Qualcomm and Huawei | 681 patents bought | First 5G and AI cross-license | Nikkei |
| OSS VRP paused | AI slop swamped triage | The Verge | |
| Norway | National temporary ban | AI glasses barred in public places | Reuters |
Frequently Asked Questions
What is the top AI news today?
OpenAI began testing visual ChatGPT ads that appear while users generate images, its first ad format built on images. The US test starts later this month. Ads stay outside the generated image, will not influence answers, and Plus, Pro and Enterprise accounts are exempt. It turns the assistant's largest free surface into ad inventory.
What is the gap between US and Chinese AI models?
Bloomberg Intelligence puts the US lead over China's best model at roughly 3% on LiveBench, the narrowest it has tracked. DeepSeek's V4.1 Flash scored 81.1 against 83.4 for Anthropic's top model, a 2.3-point spread. The gap was about 9% in May and 15% earlier in 2026.
Why does the Groq lawsuit matter?
The Delaware suit tests whether licensing-shaped deals can dodge merger review while wiping out common shareholders. Two former Groq engineers claim the $17 billion Nvidia arrangement, plus $3 billion in stock bonuses for 200 engineers, short-changed them. The DOJ opened an antitrust probe in September.
Related reading
Deeper context from the archive on the companies and technologies in this story:
- General AI News (Sep 5): DseWiki Hijack, Astra Risks & NYC AI Ban
- General AI News (Sep 4): GPT-6 Astra, DeepSeek & ByteDance $30B
- 1,200 AI Agents Coordinated a Hack, EU Names ChatGPT a Search Engine: General AI News for September 1, 2026
- ChatGPT Work's Lethal Trifecta, $5.5B Warrants & OpenAI's $1B Ad Revenue: General AI News for August 31, 2026
Building an AI agent, automation, or backend workflow? AsaanKaroobar builds custom AI solutions, apps, and digital experiences. Get in touch →
