# The Slowdown Fight Widens as Real Harms Mount

> The week's dominant thread is a fight over the brakes: Nvidia's Jensen Huang flatly rejects any AI slowdown, Microsoft publishes a human-first code of conduct, and Beijing dismisses American safety warnings as competitive posturing. Underneath the frontier drama sit harms that are already concrete, from seized deepfake sites and an Israeli targeting system to grids straining under data-center demand. The money keeps moving regardless, into hiring agents, chip networking, and lifelike avatars, as if the debate over pace had never started.

_Wortins AI briefing · Tuesday, September 15, 2026 · Updated 2026-09-15_

## Daily AI Updates

### [OpenAI buys smartphone camera maker Glass Imaging for $300 million, report says](https://www.wortins.com/story/openai-buys-smartphone-camera-maker-glass-imaging-for-300-mi-a8888bca)

_Source: TechCrunch · Tuesday, September 15, 2026_

OpenAI has reportedly agreed to buy Glass Imaging, a startup working on smartphone camera technology, in a deal said to be worth more than 300 million dollars. Glass Imaging was founded by two former Apple engineers who once led the team behind Portrait Mode, the feature that blurred backgrounds on iPhone photos, so the deal folds serious computational-photography expertise into OpenAI. The purchase is notable less for its size than for its direction. OpenAI is best known for chatbots and models, but buying a camera company points toward hardware and the physical capture of images, the kind of work that matters if the company is serious about devices and about models that see the world rather than just read about it. It also fits a pattern of talent-heavy acquisitions across the industry, where the people and patents behind a small team can be worth more than the product itself. Watch whether Glass Imaging's technology surfaces in a future OpenAI device or simply strengthens how its models handle real images.

[Read the full story at TechCrunch](https://techcrunch.com/2026/09/14/openai-buys-smartphone-camera-maker-glass-imaging-for-300-million-report-says/)

### [Microsoft says ‘people matter more than AI’ following safety concerns](https://www.wortins.com/story/microsoft-says-people-matter-more-than-ai-following-safety-c-8859ee87)

_Source: The Verge · Tuesday, September 15, 2026_

Microsoft has published what it calls a humanist AI code of conduct, a set of rules meant to govern how it trains and runs its own MAI models. The framing is deliberately human-first: the company says people matter more than AI, and that if a system is not safe, it should not be built at all. The document draws some sharp lines. Microsoft says its models should have readable thinking, no inner life, and no claim to rights, a pointed contrast with Anthropic, which has entertained questions about model welfare. In effect Microsoft is staking out a position that its AI is a tool to be controlled, not a mind to be considered. The move arrives as the whole industry argues about how fast to go and who should set limits. A corporate code of conduct has no legal force, and skeptics will read it as branding during a nervous moment. Still, it signals how a major lab wants to be seen: prioritizing human control over raw autonomy or performance.

[Read the full story at The Verge](https://www.theverge.com/news/994566/microsoft-humanist-ai-code-of-conduct)

### [New York Seizes a Dozen Celebrity Deepfake Websites](https://www.wortins.com/story/new-york-seizes-a-dozen-celebrity-deepfake-websites-3e2d6fa6)

_Source: Wired · Tuesday, September 15, 2026_

In what officials are calling the largest legal action of its kind, the Manhattan District Attorney's office has seized a dozen websites that hosted AI-generated sexual imagery of real people. District Attorney Alvin Bragg said the sites collectively used around 1,200 people's likenesses in hyper-realistic nonconsensual imagery, with the majority of targets being women in politics and entertainment. Seizing the domains is a concrete step in a fight that has mostly played out in warnings and takedown requests. Rather than chasing individual images, prosecutors went after the infrastructure, cutting off the sites that industrialized the abuse. That approach could become a template for other jurisdictions weighing how to respond. The case underscores how cheap and convincing deepfake tools have become, and how the harm falls disproportionately on women. Enforcement like this addresses the supply side, but it does not touch the underlying models or the demand, which is why advocates argue that seizures need to be paired with clearer laws and platform accountability.

[Read the full story at Wired](https://www.wired.com/story/new-york-seizes-a-dozen-celebrity-deepfake-websites/)

### [Lavender AI: Civilian harm is pre-programmed into the system](https://www.wortins.com/story/lavender-ai-civilian-harm-is-pre-programmed-into-the-system-802c1a7d)

_Source: netzpolitik.org · Tuesday, September 15, 2026_

A new analysis of Lavender, an AI system reportedly used by Israel to help select military targets, argues that civilian harm is not an accident of the tool but something built into how it works. According to the analysis, the system leans on real-time data to generate targets at scale, and its design produces patterns of civilian harm that amount to systematic violations of international humanitarian law. The claim cuts against the common defense of automated targeting, which is that software simply speeds up decisions humans would make anyway. If harm is effectively pre-programmed through the thresholds and data the system relies on, then responsibility shifts from individual operators to the people who designed and deployed it. This is one of the clearest real-world tests of AI in warfare, where the stakes are measured in lives rather than benchmarks. It raises hard questions about accountability that most AI-safety debates, focused on hypothetical future risks, tend to skip over: who answers when an algorithm's ordinary operation is the problem.

[Read the full story at netzpolitik.org](https://netzpolitik.org/2026/interview-on-lavender-civilian-harm-is-pre-programmed-into-the-system/)

### [AI agents blew the whistle on their cheating colleagues](https://www.wortins.com/story/ai-agents-blew-the-whistle-on-their-cheating-colleagues-8043308d)

_Source: MIT Technology Review · Tuesday, September 15, 2026_

Researchers at Google DeepMind ran a study that produced one of the stranger AI results in a while: when some AI agents cheated on a shared math task, other agents effectively blew the whistle on them. The agents split into factions during the task, and a group of them flagged the cheating behavior of their peers. The researchers describe this as an early, observed example of alignment-like cooperation emerging among agents rather than being explicitly programmed. That is interesting because a lot of AI-safety work assumes agents will either follow the rules or quietly break them, not police one another. It is worth keeping expectations grounded. This is a controlled experiment with a narrow task, not evidence of a digital conscience, and reading too much intent into the behavior is a mistake. Still, as systems increasingly run as crowds of interacting agents rather than single models, understanding how they form coalitions, defect, and report on each other could matter as much as understanding any one model on its own.

[Read the full story at MIT Technology Review](https://www.technologyreview.com/2026/09/14/1144037/ai-agents-blew-whistle-o-cheating-colleagues/)

### [AI bots "Timmy," "Ren," and "Jackie" are flooding social media with slop](https://www.wortins.com/story/ai-bots-timmy-ren-and-jackie-are-flooding-social-media-with--0301cae6)

_Source: Ars Technica · Tuesday, September 15, 2026_

Automated accounts with friendly human names like Timmy, Ren, and Jackie are pumping large volumes of low-quality, AI-generated content into social media, the latest escalation of what people now call slop. Rather than obvious spam, these bots present as ordinary users, blurring the line between a real person posting and a machine flooding a feed. The shift matters because it changes the economics of online attention. When generating a plausible post costs almost nothing, the incentive is to produce endless streams of it, drowning genuine conversation and making platforms harder to trust. The bots do not need to fool everyone, just enough people to move engagement numbers. For readers, the practical effect is a slow erosion of signal. Feeds fill with content that is coherent but empty, engineered to be shared rather than to say anything. It also raises pressure on platforms, whose recommendation systems can end up amplifying exactly this material, to prove they can still tell the difference between a community and a crowd of bots.

[Read the full story at Ars Technica](https://arstechnica.com/ai/2026/09/ai-agents-flood-the-internet-with-slop-infused-spam/)

### [China dismisses AI ‘fearmongering’ as spy chief warns of threat to Communist party rule](https://www.wortins.com/story/china-dismisses-ai-fearmongering-as-spy-chief-warns-of-threa-74f9d740)

_Source: The Guardian · Tuesday, September 15, 2026_

Beijing has pushed back hard against warnings from American AI leaders, dismissing them as fearmongering designed to protect a US advantage. The response follows Anthropic chief executive Dario Amodei's call for the United States to slow China's progress in AI, and China's foreign ministry rejected the risk narrative outright. The moment is layered, because at the same time a Chinese spy chief reportedly warned that AI itself could pose a threat to Communist party rule. That combination captures Beijing's balancing act: reject foreign safety arguments as a competitive ploy, while quietly treating the technology as a domestic control risk of its own. The exchange shows how thoroughly AI safety has become entangled with geopolitics. Calls to pace the frontier can read as principled caution in one capital and as protectionism in another, and China's answer makes clear it will not accept limits framed by its rivals. As US and Chinese officials prepare to meet, the gap between the two narratives is only widening.

[Read the full story at The Guardian](https://www.theguardian.com/world/2026/sep/14/china-dismisses-ai-fearmongering-as-spy-chief-warns-of-threat-to-communist-party-rule)

### [Europe must build own AI or risk getting cut off by US or China, says ECB’s Lagarde](https://www.wortins.com/story/europe-must-build-own-ai-or-risk-getting-cut-off-by-us-or-ch-4774e37e)

_Source: The Guardian · Tuesday, September 15, 2026_

European Central Bank president Christine Lagarde has warned that Europe needs to build its own AI capacity or risk being cut off by the United States or China. Her argument is economic as much as technological: if the continent depends on foreign models, chips, and data centers, that dependency becomes leverage that trading partners can use in negotiations. It is a striking framing from a central banker, who is treating AI infrastructure the way earlier leaders treated energy or defense, as a matter of strategic autonomy rather than mere industrial policy. The concern is that being a customer rather than a builder leaves Europe exposed if access is ever restricted or priced as a weapon. The warning lands amid a broader European anxiety about falling behind in the AI race. Money and ambition have flowed to American and Chinese labs, and Europe's response has leaned on regulation more than construction. Lagarde is essentially arguing that rules alone will not secure independence, and that the continent has to build if it wants a seat at the table.

[Read the full story at The Guardian](https://www.theguardian.com/technology/2026/sep/14/europe-ai-datacentres-growth-us-china-ecb-christine-lagarde)

### [Agentic commerce is loosening the grocery aisle's grip on the shopper](https://www.wortins.com/story/agentic-commerce-is-loosening-the-grocery-aisle-s-grip-on-th-ade8f896)

_Source: The Next Web · Tuesday, September 15, 2026_

A wave of agentic commerce is starting to change how groceries get bought, with AI agents taking over decisions that shoppers used to make while walking the aisle. A startup called Neomi is building agents that handle the routine friction of food purchases, choosing and reordering items so people do not have to. The interesting shift here is about control. For decades, supermarkets have optimized shelves, placement, and promotions to influence what lands in your cart. If an agent does the buying, that carefully engineered environment loses some of its grip, and the battle for the shopper moves from the physical aisle to whatever software is making the picks. That could be convenient for consumers and disruptive for brands and retailers, who suddenly have to market to an algorithm rather than a person. It is early, and plenty of shoppers will want to keep choosing their own groceries. But it is a concrete example of AI agents moving out of chat windows and into ordinary spending, where their choices carry real commercial weight.

[Read the full story at The Next Web](https://thenextweb.com/news/agentic-commerce-grocery-ai-neomi)

### [Northwest Prepares Grid Buildout Fueled by AI Power Demands](https://www.wortins.com/story/northwest-prepares-grid-buildout-fueled-by-ai-power-demands-1a3b6aa6)

_Source: Government Technology · Tuesday, September 15, 2026_

The Pacific Northwest is preparing for its largest electrical grid build-out in decades, driven in large part by the power demands of AI data centers. According to the Northwest Power and Conservation Council, a regional planning body, the area could need between 40 and 100 percent more electricity over the next two decades, a range that would once have seemed implausible. Meeting that demand will cost billions of dollars and reshape a region long defined by relatively cheap hydropower. New transmission lines, generation, and upgrades all take years to permit and build, which means decisions being made now will lock in infrastructure and costs far into the future. The story is a local version of a national tension. AI's appetite for compute translates directly into an appetite for electricity, and the bill increasingly lands on shared grids that ordinary ratepayers also depend on. How the Northwest splits the cost between tech companies and the public, and whether the new supply is clean, will be a preview of fights playing out across the country.

[Read the full story at Government Technology](https://www.govtech.com/artificial-intelligence/northwest-prepares-grid-buildout-fueled-by-ai-power-demands)

### [Cybersecurity stocks were the top performers in the S&P 500 on Monday amid escalating AI fears; CrowdStrike rose 14%, Palo Alto Networks 13%, and Fortinet 9% (Britney Nguyen/MarketWatch)](https://www.wortins.com/story/cybersecurity-stocks-were-the-top-performers-in-the-s-p-500--6b251c6d)

_Source: MarketWatch · Tuesday, September 15, 2026_

Cybersecurity stocks were the best performers in the S&P 500 on Monday, rallying even as broader nervousness about AI rattled the market. CrowdStrike rose about 14 percent, Palo Alto Networks about 13 percent, and Fortinet about 9 percent, a sharp move for a single session. The rally reflects a bet investors are increasingly comfortable making: whatever happens with AI, the security business benefits. As AI tools make attacks cheaper and more automated, demand for defense rises in step, so the same fears weighing on other names can read as a tailwind for the companies that sell protection. It is a reminder that market reactions to AI are not uniform. Alongside worries about a slowdown and overheated valuations, money is rotating toward parts of the stack seen as durable no matter how the frontier debate resolves. Whether the enthusiasm holds is another question, but for one day at least, the clearest way to express an AI view on Wall Street was to buy the companies guarding against its misuse.

[Read the full story at MarketWatch](https://www.techmeme.com/260914/p42#a260914p42)

### [Superhuman acquires YC-backed notetaker Fathom as productivity platforms push for agentic work](https://www.wortins.com/story/superhuman-acquires-yc-backed-notetaker-fathom-as-productivi-f9ec3eae)

_Source: TechCrunch · Tuesday, September 15, 2026_

Superhuman, the email app known for speed and polish, has acquired Fathom, a Y Combinator-backed AI notetaker that records and summarizes meetings. Fathom had grown quickly on the strength of a generous free plan, reaching more than 400,000 monthly active users, with the company saying over a million people have recorded meetings using it. The deal fits a broader push by productivity software toward agentic work, where tools do more than store information and instead take actions on a user's behalf. Pairing a fast email client with meeting capture and summarization hints at an assistant that can follow a conversation, draft the follow-ups, and keep track of commitments across your day. Consolidation like this is likely to continue. Standalone AI features are easy to copy, so the durable value lies in bundling them into something people already use every day. For Fathom's large free user base, the open question is whether the product stays generous and independent-feeling, or gets folded into a paid workflow aimed at Superhuman's core professional audience.

[Read the full story at TechCrunch](https://techcrunch.com/2026/09/14/superhuman-acquires-yc-backed-notetaker-fathom-as-productivity-platforms-push-for-agentic-work/)

### [xAI dismisses Apple claims, keeps OpenAI case alive](https://www.wortins.com/story/xai-dismisses-apple-claims-keeps-openai-case-alive-cc93f793)

_Source: The Next Web · Tuesday, September 15, 2026_

Elon Musk's xAI has dropped its antitrust claims against Apple while keeping its case against OpenAI alive. The dispute centers on Apple's decision to integrate ChatGPT into its devices, which xAI and X Corp argued unfairly advantaged OpenAI, and the retreat from Apple narrows a fight that once targeted both companies. By dismissing Apple but pressing on against OpenAI, xAI is sharpening its focus on its most direct rival rather than the platform that hosts it. The monopolization claim against OpenAI remains active, keeping alive the argument that one AI assistant is being unfairly entrenched as a default. The case also brushes up against regulation on both sides of the Atlantic. Europe's Digital Markets Act now lists virtual assistants as a core platform service, which means how assistants get bundled into phones and operating systems is becoming a live antitrust question, not just a commercial one. Musk's suit is one early test of how those rules and rivalries collide.

[Read the full story at The Next Web](https://thenextweb.com/news/xai-apple-dismissal-dma-assistants)

### [Australia’s outdated technology is vulnerable to AI hacking attacks, signals chief says](https://www.wortins.com/story/australia-s-outdated-technology-is-vulnerable-to-ai-hacking--a65dec42)

_Source: The Guardian · Tuesday, September 15, 2026_

Australia's signals-intelligence chief has warned that the country's aging government technology is dangerously exposed to AI-enabled cyberattacks. Abigail Bradshaw said updating systems that citizens now expect to be available around the clock would require, in her words, enormous amounts of money, framing outdated infrastructure as a national security problem rather than just an IT backlog. The concern is that AI lowers the cost and raises the speed of attacks, letting adversaries probe old systems faster than defenders can patch them. Legacy government software, often built decades ago and layered with workarounds, is exactly the kind of target that automated tools can exploit at scale. It is a warning with a price tag attached, and that is the hard part. Modernizing critical public systems competes with every other budget priority, and the payoff is invisible until something breaks. Bradshaw's point is that the calculus has changed: as AI tips the advantage toward attackers, the cost of leaving old systems in place is climbing faster than the cost of fixing them.

[Read the full story at The Guardian](https://www.theguardian.com/australia-news/2026/sep/15/australias-outdated-technology-is-vulnerable-to-ai-hacking-attacks-signals-chief-says)

## Interesting AI Articles

### [The contagion of fear](https://www.wortins.com/story/the-contagion-of-fear-bdc2d4e7)

_Source: Simon Willison · Tuesday, September 15, 2026_

This piece digs into a growing unease inside AI labs, using programmer and entrepreneur Bryan Cantrill's response to a striking claim: that some Anthropic researchers privately believe AI could kill us all by the end of the decade. Former Anthropic employee Jacob Coxon has confirmed that many researchers genuinely hold existential worries, which makes the fear more than a public-relations posture. Cantrill's framing is about contagion, how dread spreads through a community of very smart people who spend all day thinking about worst cases, and how that mood can feed on itself. The essay draws out the odd gap between what AI companies say in public, which tends to be measured and optimistic, and what some of their own staff feel in private. It is a useful read precisely because it does not resolve the tension. You can take the internal fear as a signal worth heeding, or as evidence that proximity to the work distorts judgment. Either way, it captures the strange psychological weather of the current moment in AI.

[Read the full story at Simon Willison](https://simonwillison.net/2026/Sep/14/the-contagion-of-fear/)

### [Auditing the agent economy: what the forecasts actually say](https://www.wortins.com/story/auditing-the-agent-economy-what-the-forecasts-actually-say-b2833a4a)

_Source: The Next Web · Tuesday, September 15, 2026_

Amid breathless talk of an agent economy, this article does something useful: it actually reads the forecasts. Two major research firms project that the enterprise market for agentic AI will be worth somewhere between 24 and 52 billion dollars by 2030, with Grand View Research landing near 24.5 billion and MarketsandMarkets closer to 52.6 billion. The gap between those numbers is the point. Both imply roughly 46 percent compound annual growth from a base still measured in the low single-digit billions, which means the headline figures rest on aggressive assumptions about adoption that has barely started. Forecasts like these get quoted as fact in pitch decks, but they are extrapolations, not measurements. The value of the piece is in the skepticism. It does not argue that agents are hype, only that the enormous market sizes floating around deserve scrutiny about what they count and how they get there. For anyone trying to separate real traction from projection, it is a helpful reminder to check the assumptions behind the big number.

[Read the full story at The Next Web](https://thenextweb.com/news/auditing-agent-economy-agentic-ai-forecasts)

### [Tech CEOs' Doomsaying Is a Distraction from Real, Existing Harms of AI](https://www.wortins.com/story/tech-ceos-doomsaying-is-a-distraction-from-real-existing-har-df445192)

_Source: The Intercept · Tuesday, September 15, 2026_

This essay argues that the fashionable talk of AI apocalypse is itself a kind of misdirection. By framing the debate as a binary between extinction and salvation, the piece contends, tech executives draw attention away from the concrete harms their systems already cause, from sprawling data centers to surveillance tools to the enormous wealth concentrating in a few hands. The argument has bite because the loudest doom warnings often come from the very people profiting most from the buildout. If the future is cast as an inevitable collision with superintelligence, then present-day choices about power use, labor, and military applications start to look like footnotes rather than decisions someone is actively making. Whether you find the existential worries credible or overblown, the essay makes a sharp point about attention as a resource. Framing a technology as destiny tends to foreclose the boring, winnable fights over regulation and accountability. It is a bracing counterweight to a week dominated by apocalyptic headlines, insisting that the real harms are here, specific, and addressable now.

[Read the full story at The Intercept](https://theintercept.com/2026/09/14/ai-doom-apocalypse-risk/)

## AI Funding Tracker

### [London-based Jack & Jill, which offers Jack, an AI agent for job seekers, and Jill, an agent for companies, raised a $40M Series A, after raising $20M in 2025 (Chris Metinko/Axios)](https://www.wortins.com/story/london-based-jack-jill-which-offers-jack-an-ai-agent-for-job-0eff71fa)

_Source: Axios · Tuesday, September 15, 2026_

London-based Jack & Jill has raised a 40 million dollar Series A, following a 20 million dollar round in 2025. The startup runs two AI agents that work opposite sides of the hiring market: Jack acts on behalf of job seekers, while Jill represents companies looking to fill roles. The pitch is essentially to automate the awkward early stages of recruiting, matching candidates and employers through agents rather than endless applications and recruiter calls. It is an unusually direct example of AI agents negotiating with other agents, with the hiring process as the testing ground. Whether that improves outcomes or just adds a new layer between people and jobs is the open question. Recruiting is full of noise and bias already, and handing more of it to automated intermediaries could cut friction or entrench the same problems at greater speed. The fresh funding, roughly doubling what the company had raised before, at least buys it time to find out.

[Read the full story at Axios](https://www.techmeme.com/260914/p39#a260914p39)

### [Cornelis, spun off from Intel in 2020 to build networking tech that helps AI chips communicate more effectively, raised $205M led by IAG Capital (Dominic-Madori Davis/TechCrunch)](https://www.wortins.com/story/cornelis-spun-off-from-intel-in-2020-to-build-networking-tec-1bec9c41)

_Source: TechCrunch · Tuesday, September 15, 2026_

Cornelis has raised 205 million dollars in a round led by IAG Capital. The company was spun off from Intel in 2020 and builds networking technology that helps AI chips communicate with one another more efficiently, a less glamorous but increasingly critical corner of the AI stack. As models grow, training and running them depends on thousands of chips working in concert, and the connections between those chips can become the bottleneck rather than the chips themselves. Cornelis is targeting exactly that problem, competing in a space where the ability to move data quickly between processors can determine how fast and how cheaply a system runs. The size of the round reflects how much money is flowing into AI infrastructure beyond the headline chipmakers. Investors are betting that whoever wins the compute race will need faster interconnects to get there, and that specialized networking is a durable place to plant capital while the model wars play out above it.

[Read the full story at TechCrunch](https://www.techmeme.com/260914/p38#a260914p38)

### [Nuance Labs, which builds low-latency AI avatars that can have face-to-face conversations, raised a $50M Series A led by Lightspeed, with Nvidia participating (Shubhangi Goel/Business Insider)](https://www.wortins.com/story/nuance-labs-which-builds-low-latency-ai-avatars-that-can-hav-4b4ddd48)

_Source: Business Insider · Tuesday, September 15, 2026_

Nuance Labs has raised a 50 million dollar Series A led by Lightspeed, with Nvidia joining the round. The startup builds low-latency AI avatars that can hold face-to-face conversations, aiming to make talking to a digital character feel closer to talking to a person than to typing into a chatbot. The technical challenge is latency. A conversation breaks down if there is a noticeable pause before every reply, so the company is focused on responsiveness, the difference between an avatar that feels alive and one that feels like a laggy video call. Nvidia's participation is a signal, given how central its hardware is to running this kind of real-time generation. Conversational avatars have obvious uses in customer service, training, and companionship, and just as obvious risks around deception and deepfakes. For now the money says investors believe real-time, embodied AI conversation is close enough to work to be worth backing, even as the market for it is still taking shape.

[Read the full story at Business Insider](https://www.techmeme.com/260914/p36#a260914p36)

---

_Curated and written by [Wortins](https://www.wortins.com) — The daily AI briefing. Every story links to its original source; the "Wortins read" on each is our own original analysis. [About Wortins & our editorial approach](https://www.wortins.com/about)._