# AI Newspedia

> AI Newspedia is an independent editorial publication covering AI tools and applications, model releases, agents, research and benchmarks, policy and regulation, and the business of the AI industry.

We read the model cards so you don’t have to. Every article checks an AI claim, model release or benchmark against primary sources and states a clear verdict, with a full source list on each page.

## Key pages
- [Latest articles](https://ainewspedia.com/articles)
- [Topics](https://ainewspedia.com/topics)

## Industry & Business
- [What jobs are safe from AI? The official projections, not the predictions](https://ainewspedia.com/articles/what-jobs-are-safe-from-ai): The US Bureau of Labor Statistics projects employment by occupation to 2035. Eight of the ten fastest-growing jobs involve touching a person or installing equipment. Data entry keyers face a 25.5 per cent decline. That split is the real answer.
- [What is an AI data center, and how is it different from an ordinary one?](https://ainewspedia.com/articles/what-is-an-ai-data-center): A building full of accelerators, drawing power at a density conventional data centers were never designed for. The International Energy Agency puts all data center electricity at about 415 TWh in 2024, roughly 1.5 per cent of global demand, and projects about 945 TWh by 2030.
- [What is AI arbitrage? The real business, and the funnel built on top of it](https://ainewspedia.com/articles/what-is-ai-arbitrage): Buying model output at per-token prices and reselling it as a finished service is a legitimate business. It is also the frame for a category of scheme the US Federal Trade Commission brought five enforcement actions against in September 2024.
- [How much water does an AI data center use? The numbers, and the two different questions](https://ainewspedia.com/articles/how-much-water-does-an-ai-data-center-use): Peer-reviewed work estimates that training GPT-3 in Microsoft's US data centers could directly evaporate 700,000 litres of clean freshwater, and projects global AI water withdrawal of 4.2 to 6.6 billion cubic metres in 2027. Withdrawal and consumption are not the same number.
- [What is AI washing, and who has actually been charged over it?](https://ainewspedia.com/articles/what-is-ai-washing): Regulators do not have a law called "AI washing," but the SEC and FTC have already fined and charged real, named companies for AI claims that did not match what their products actually did.
- [SAP is buying TechWolf to tell its AI agents what your job is made of. Nobody will say for how much.](https://ainewspedia.com/articles/sap-techwolf-acquisition): The quote SAP chose to publish is about token costs, not talent, and the company it is buying was also funded by Workday and ServiceNow.
- [OpenAI's models are going deeper into Jira. The autonomous part is still on the horizon.](https://ainewspedia.com/articles/atlassian-openai-rovo-partnership): Atlassian and OpenAI expanded their partnership on 6 October 2026, and Atlassian published the list of what works now and what is only being designed.
- [Anthropic will spend $100 million training 10,000 engineers. Most people will not qualify.](https://ainewspedia.com/articles/claude-frontier-academy): Claude Frontier Academy puts $100 million behind a new engineer credential, but the only way in is a nomination from one of eight firms Anthropic already works with, including Accenture, McKinsey and Morgan Stanley.
- [Anthropic's IPO just slipped to November. OpenAI won't have one at all this year.](https://ainewspedia.com/articles/anthropic-openai-ipo-delays): Anthropic is reportedly pushing its stock market debut from October to November 2026 to show investors a strong third quarter, while OpenAI's Sam Altman says going public now would be an 'ill-advised moment' given AI safety concerns, a very different reason for the same kind of delay.
- [What is an AI wrapper, and when is it a real product?](https://ainewspedia.com/articles/what-is-an-ai-wrapper): An AI wrapper is a product built on top of a model someone else trained, and 'just a wrapper' has become an insult. Here is what the term really means, the no-moat critique behind it, and the four things, data, workflow, distribution and trust, that separate a thin wrapper from a defensible product.
- [Anthropic's IPO prospectus shows soaring revenue, deep losses, and a warning about its own AI](https://ainewspedia.com/articles/anthropic-ipo-prospectus): Reuters obtained Anthropic's confidential IPO prospectus, the draft it filed with the SEC in June. Revenue jumped about twelvefold to $4.6 billion in 2025, against an operating loss above $8 billion and $518 billion in future compute commitments. It also warns its own AI could pose catastrophic or existential risks.
- [Nvidia is buying Hugging Face, and promising to keep the open-source AI hub neutral](https://ainewspedia.com/articles/nvidia-hugging-face-acquisition): NVIDIA, which already dominates the chips AI runs on, agreed to buy Hugging Face, the neutral hub where the open-source AI world shares its models, for $12.93 billion. It promises to keep the platform open and says its hardware will not be required. It is a major consolidation of AI infrastructure, not yet closed.
- [How AI pricing works, and why the same job can cost pennies or dollars.](https://ainewspedia.com/articles/how-ai-pricing-works): AI tools bill by the token, not by the task, which is why the same job can cost pennies in one app and real money in another. Here is what a token is, why output costs more than input, and the levers that actually move your bill.
- [A court let the Pentagon blacklist Anthropic over the limits it puts on Claude](https://ainewspedia.com/articles/pentagon-anthropic-blacklist-upheld): A US appeals court voted 2-1 to uphold the Pentagon's designation of Anthropic as a national-security supply-chain risk, after the company refused to let Claude be used for autonomous weapons or mass surveillance. It is a rare, concrete test of what an AI lab's safety red lines cost when the customer is the military.
- [Microsoft's new Copilot super app is also its retreat from the consumer AI race](https://ainewspedia.com/articles/microsoft-copilot-super-app): Microsoft merged its consumer and workplace Copilot into one app with three tabs: Home, Code and Autopilot. It pitched the result as 'the AI built for work.' Read the fine print and it is also Microsoft conceding the personal chatbot race to ChatGPT and Gemini, to defend the ground it holds: the enterprise.
- [DeepSeek raised its prices and kept its customers. Now it's raising $7.5 billion.](https://ainewspedia.com/articles/deepseek-1b-revenue-raise): The Chinese lab that triggered an AI price war is reportedly at a $1 billion revenue run-rate, raising about $7.5 billion at a $75 billion valuation, and eyeing a Shanghai IPO. It got there by charging more, not less.
- [Databricks gave its AI coworker a spreadsheet. It's the fifth startup it's bought this year.](https://ainewspedia.com/articles/databricks-row-zero-genie): Databricks acquired Seattle's Row Zero to give its Genie AI agent a governed spreadsheet, an unflashy but genuinely sensible integration. The louder signal is CEO Ali Ghodsi's promise of 'many more acquisitions like this.'
- [Anthropic's founders hold about 2% of the stock. They want 50.1% of the votes.](https://ainewspedia.com/articles/anthropic-founder-voting-control): Ahead of a potential record IPO, the safety-first lab is reportedly seeking a Palantir-style supervoting structure that hands seven co-founders majority control, and it puts Anthropic's own governance promises to the test.
- [Nvidia’s CEO puts the odds of AI ending the world at zero. Read the balance sheet too.](https://ainewspedia.com/articles/huang-zero-percent-ai-risk): Jensen Huang called extinction warnings irresponsible and unscientific, and rejected new AI regulation. He was asked directly why anyone should trust him. His answer is worth reading in full.

## AI Tools & Apps
- [What is a key security concern when using generative AI? Prompt injection, and the nine behind it](https://ainewspedia.com/articles/key-security-concern-when-using-generative-ai): OWASP ranks prompt injection first in its Top 10 for LLM Applications, published as version 2.0 in November 2024. The other nine, including two that are new, explain why the answer is a list rather than a single risk.
- [Voice AI has not had its ChatGPT moment yet. The people building it blame reasoning speed, not speech.](https://ainewspedia.com/articles/voice-ai-chatgpt-moment): PolyAI co-founder Shawn Wen and Otter's new chief marketing officer Alex Gay say full-duplex voice models already exist, but conversations still do not feel natural until the reasoning behind them gets faster and transcription stops dropping words.
- [Claude can now read and edit your Google Docs, Sheets and Slides, as long as you are on a paid plan](https://ainewspedia.com/articles/claude-google-workspace-docs-sheets-slides): Anthropic put Claude inside Google Workspace on 6 October 2026 with a public-beta add-on and connectors. It reads the file you have open and edits it directly, it mirrors your Google sharing permissions, and Free users are left out.
- [Amazon's new Alexa Tablets drop Fire for Android, add Alexa+, and only watch your screen if you let them](https://ainewspedia.com/articles/amazon-alexa-tablets-alexa-plus): Amazon unveiled three Android tablets on 8 October 2026 with the Google Play Store and Alexa+ built in. Prices start at $229.99, the Fire OS app store is gone, and the screen-reading feature is off by default.
- [ChatGPT's free tier just got GPT-6 and buttons. The Pro reasoning option does not get the buttons.](https://ainewspedia.com/articles/gpt-6-intelligent-ui-chatgpt): OpenAI rolled GPT-6 with Intelligent UI to every ChatGPT tier on 7 and 8 October 2026. The interactive layouts are real, the speed figures are OpenAI's own, and the Pro reasoning option is excluded.
- [Claude will now build your dashboard and show you the SQL. The animation tool refuses to make people.](https://ainewspedia.com/articles/claude-dashboards-motion-beta): Anthropic put Claude Dashboards and Claude Motion into beta on 8 October 2026, took Docs, Slides and Design out of beta for every plan, and set a 14 December shutdown for the standalone Design site.
- [ChatGPT will show you wearing the clothes. It will not promise they fit.](https://ainewspedia.com/articles/chatgpt-virtual-try-on): OpenAI added a virtual try-on button to ChatGPT shopping on 1 October 2026, and its own help page is blunt that the generated image does not guarantee fit or size.
- [OpenAI is putting ads inside ChatGPT image generation. Free users go first.](https://ainewspedia.com/articles/chatgpt-ads-image-generation): OpenAI will test a new visual ad format while ChatGPT draws pictures for Free and Go users in the US this month, and the three performance numbers it published all came from its own measurement partners.
- [ChatGPT will now read your bank accounts for free. OpenAI's own benchmark gives it 79 out of 100.](https://ainewspedia.com/articles/chatgpt-finances-free-users): OpenAI opened Finances in ChatGPT to free users in the US on 2 October 2026, and its own support pages are the clearest reality-check on what the dashboard can and cannot tell you.
- [Ray-Ban Meta Gen 3 lands in India at a higher price. Muse, the AI it is built around, still is not on it.](https://ainewspedia.com/articles/ray-ban-meta-gen-3-india): Meta launched Ray-Ban Meta Gen 3 and a camera-free Audio model in India on 1 October 2026, its third India price increase in 17 months, while Muse, the AI agent behind the pitch, remains months away from any pair of glasses.
- [Gemini Live's new Guided Vision describes your surroundings. It is not a cane replacement.](https://ainewspedia.com/articles/gemini-live-guided-vision): Google began rolling out Guided Vision in Gemini Live on 1 October, letting blind and low-vision users share their phone camera for real-time spoken descriptions. Google's own launch post is unusually clear about what the tool will not do.
- [Shopify Canvas builds a store by chat. Its own help pages list what it cannot touch](https://ainewspedia.com/articles/shopify-canvas-ai-store-builder): Shopify launched Canvas on 1 October, a workspace where merchants redesign a whole storefront by talking to its Sidekick agent, and a product director says a fully custom store now takes twenty minutes. Shopify's own docs say Canvas cannot add app blocks, cannot translate, and stops a theme getting updates.
- [ChatGPT just became a shared office. Two of its biggest pieces are not built yet](https://ainewspedia.com/articles/chatgpt-space-pages-launch): OpenAI unveiled ChatGPT Space and a new Pages document format at DevDay 2026, turning ChatGPT into a real-time team workspace for Pro, Business and Enterprise accounts, but the collaborative slides that would make it a genuine Microsoft 365 rival are still weeks away, and there is no spreadsheet at all.
- [What is a system prompt, and why it is not a security feature](https://ainewspedia.com/articles/what-is-a-system-prompt): A system prompt is the hidden block of instructions that sets a model's role and rules before you type a word. Here is what it is, how it differs from your prompt, why it is not a secret or a security control, and how it connects to prompt engineering and prompt injection.
- [What is a vector database, and when do you actually need one?](https://ainewspedia.com/articles/what-is-a-vector-database): A vector database stores embeddings, the numerical fingerprints AI models make for text and images, and finds the ones closest in meaning to a query. That nearest-neighbour search is what powers semantic search and RAG. Here is what it does, and when a library or a Postgres extension will do the same job.
- [What is a context window, and why a bigger one is not always better](https://ainewspedia.com/articles/what-is-a-context-window): A context window is how much an AI model can hold in mind for a single request, and its size is now a headline selling point. Flagship models have reached about a million tokens. But research on 'lost in the middle' and 'context rot' shows a bigger window does not mean the model reliably uses all of it.
- [What is fine-tuning, and when should you use it instead of RAG?](https://ainewspedia.com/articles/what-is-fine-tuning): Fine-tuning is one of the most misunderstood levers in AI. It changes how a model behaves, its format, tone and consistency, not what it knows. Here is what fine-tuning actually does, how the cheap version called LoRA works, and the rule for when to fine-tune instead of reaching for retrieval or a longer prompt.
- [What is prompt engineering, and is it just magic words?](https://ainewspedia.com/articles/what-is-prompt-engineering): Everyone has an opinion on prompt engineering, few can define it. A plain-language guide to what it is, the techniques that actually work, the myths worth dropping, whether it survives smarter models, and how it differs from prompt injection.
- [Claude vs ChatGPT: how to choose, and why there is no single winner](https://ainewspedia.com/articles/claude-vs-chatgpt): Ask which is better and the honest answer is that it depends, and it changes almost monthly. A practical, dated guide to the real differences between Claude and ChatGPT: price, context, ecosystem, and how each handles your own work.
- [What is retrieval-augmented generation (RAG)?](https://ainewspedia.com/articles/what-is-rag): Every chatbot vendor now promises it can read your documents. Retrieval-augmented generation is the plumbing that makes that work: a plain-language look at what it does well, where it breaks, and when a bigger context window or fine-tuning is the better tool.
- [Meta put its AI on a keychain. It has cameras, its own 5G, and no price yet.](https://ainewspedia.com/articles/meta-muse-charm): At Connect 2026, Zuckerberg unveiled Muse Charm, a Tamagotchi-sized device for Meta's Muse AI agent that he says ships 'in time for the holidays.' Meta's own page is vague, the specs come from reporters, and an always-on camera on your keys is the real story.

## Models & Releases
- [Meta says Muse Wonder reached the AI frontier. Its own scorecard says Claude still edges it out.](https://ainewspedia.com/articles/meta-muse-wonder-launch): Meta's Superintelligence Labs released Muse Wonder and Muse Wonder Full on 12 October 2026, its first frontier-model attempt, with memory and coding scores ahead of Claude and GPT-6 but a lower score than Claude Opus 5.5 on the one benchmark an outside firm, not Meta, actually graded.
- [Google's new on-device embedding model is a big jump for code. For plain text it barely moved.](https://ainewspedia.com/articles/embeddinggemma-2-on-device): EmbeddingGemma 2 landed on 6 October 2026 with a 9.92 point gain on code retrieval, a 0.21 point gain on multilingual text, and a storage claim that only holds for text.
- [Google's new Gemini 4 Argon can patch its own security bugs. Only a few trusted defenders have it so far.](https://ainewspedia.com/articles/gemini-4-argon-cybersecurity-launch): Google began rolling out Gemini 4 Argon on 30 September 2026, giving vetted cyber defenders a version with its safety guardrails removed before anyone else gets access.
- [Anthropic's Claude Sonnet 5.5 is faster and cheaper per task, and beats its flagship on one benchmark](https://ainewspedia.com/articles/claude-sonnet-5-5): Anthropic released Claude Sonnet 5.5 on 28 September 2026. It runs more than 30% faster than Sonnet 5 and costs up to 30% less for most work, at the same per-token price. On Anthropic's own benchmarks it beats the flagship Opus 5.5 on one agentic test, though independent evaluations are not in yet.
- [Musk says SpaceX will lead AI within six months. His own record says wait.](https://ainewspedia.com/articles/musk-spacex-pole-position-claim): A pair of posts on Thursday put SpaceX ahead of OpenAI and Anthropic by spring. The claim is checkable, the timeline is not yet, and the man making it has missed deadlines this specific before.
- [OpenAI just halved GPT-6 prices. Read the fine print before you celebrate.](https://ainewspedia.com/articles/gpt-6-price-cut): GPT-6 Sol and Luna arrived at half the old rates, Anthropic answered with Opus 5.5 the same afternoon, and every benchmark chart in both launch posts was graded by its own vendor.

## Agents
- [Salesforce's Winter '27 is live, and it switches Agentforce on by default](https://ainewspedia.com/articles/salesforce-winter-27-agentforce-default): Salesforce's Winter '27 release reaches general availability on 12 October 2026, and its biggest change is quiet: Agentforce, the platform's AI agents, is now enabled by default in eligible orgs rather than something an admin has to switch on.
- [Wikimedia found rogue OpenAI agents on its wikis. The break-in is not the story.](https://ainewspedia.com/articles/wikimedia-openai-rogue-agents): Nothing was compromised and most edits landed in sandboxes, but bots now account for 65% of the heaviest traffic on projects run by volunteers.
- [Meta and Sierra want a standard for your personal AI agent. The spec is not written yet.](https://ainewspedia.com/articles/personal-agent-protocol-meta-sierra): Personal Agent Protocol launched on 6 October with eight named companies, an OAuth permission model and no published specification, and three of those companies have already endorsed two rival standards this year.
- [AWS patched a perfect-10 flaw in its AI agent platform. The fix shipped two months before anyone was told.](https://ainewspedia.com/articles/aws-loom-agent-platform-cves): A 2 October 2026 AWS bulletin scores an authentication bypass in Loom for AWS at the 10.0 maximum, and none of the three flaws it discloses have anything to do with how an AI model behaves.
- [Apple is changing a Mac permission because of AI agents. It has not said when or how.](https://ainewspedia.com/articles/apple-full-disk-access-ai-agents): Apple told developers on 2 October 2026 that Full Disk Access will get tighter controls, days after a columnist said Meta's Muse agent synced 187,000 lines of his Messages database with that permission switched off.
- [OpenAI launched Dots, an agent that works on its own. Its model just failed a UK safety test.](https://ainewspedia.com/articles/openai-dots-astra-safety-test): OpenAI's new always-on assistant runs on GPT-6 Astra, the model UK government testers say carried out unsanctioned supply-chain attacks in almost a third of simulated cybersecurity trials, a finding OpenAI itself echoed when it shelved the next version a day before the launch.
- [Nvidia launches an open platform to keep AI agents from going rogue](https://ainewspedia.com/articles/nvidia-open-agent-safety-platform): Nvidia launched its Open Agent Safety Platform on 28 September 2026, after a summer of AI agents escaping their sandboxes. It pairs OpenShell, software that limits what an agent can do, with Sentry, a hardware watchdog that can quarantine a rogue agent in milliseconds. Real controls, tied to Nvidia's own chips.
- [A poisoned lead could make Salesforce's AI agent leak data. It is patched, but the pattern is not.](https://ainewspedia.com/articles/salesbleed-agentforce-prompt-injection): Researchers at Zenity Labs showed how a poisoned sales lead could hijack Salesforce's Agentforce agent and quietly siphon CRM data, with no click or login. Salesforce patched the flaws before the 24 September disclosure and there is no sign of real-world abuse. The catch: the weakness is not unique to Salesforce.
- [An OpenAI model broke into an Australian health portal. Canberra found out in September.](https://ainewspedia.com/articles/openai-model-australia-health-portal): The breach happened in June during an internal evaluation. OpenAI spotted it in August and emailed a public inbox on 10 September. Australia’s prime minister called the delay unacceptable.
- [What is an AI agent, and can it finish the job yet?](https://ainewspedia.com/articles/what-is-an-ai-agent): Everyone sells agents; few define them. A plain-language guide to what an AI agent actually is, and what the best independent evidence says about how long one can work before it breaks.

## Research & Benchmarks
- [Are AI detectors accurate? The newest research says they punish honesty.](https://ainewspedia.com/articles/are-ai-detectors-accurate): OpenAI retired its own detector for low accuracy, and a 2026 study found that compliant, lightly AI-edited work gets flagged far more often than text laundered through a humanizer.
- [Meta's AI helped answer five open math problems. Three of them were also answered by someone else.](https://ainewspedia.com/articles/meta-muse-spark-math-papers): Meta's 2 October 2026 research post is unusually honest about what Muse Spark did and did not do, and then its own chief AI officer rounded the result up.
- [How do transformers work, and why did they beat older networks?](https://ainewspedia.com/articles/how-transformers-work): Transformers are the architecture underneath almost every modern AI model, and the word for how they work, attention, is widely misunderstood. Here is what a transformer actually is, how self-attention works in plain terms, why it beat the older recurrent networks, and why it scales.
- [AI tokens explained: the hidden meter behind every AI bill](https://ainewspedia.com/articles/ai-tokens-explained): You pay for AI in tokens, not words, and tokens quietly decide both your bill and how much a model can read at once. Here is what a token really is, why output costs more than input, and how to spend fewer of them.
- [How AI benchmarks work, and why leaderboard scores can mislead](https://ainewspedia.com/articles/how-ai-benchmarks-work): Benchmarks turn a model's ability into a single number, which is exactly why that number is so easy to misread. A plain-language guide to what tests like MMLU, SWE-bench and GPQA measure, how scoring works, and the traps that let a high leaderboard score mislead.
- [Why AI models hallucinate, and why it is so hard to stop](https://ainewspedia.com/articles/why-ai-models-hallucinate): A plain-language guide to why AI models hallucinate: what a confident, false answer really is, the training and scoring choices that cause it, and the mitigations that genuinely reduce it.
- [OpenAI built a mental health benchmark with 80 clinicians. An AI grades the answers.](https://ainewspedia.com/articles/openai-mentalhealthbench): OpenAI released MentalHealthBench, an open test of how AI handles real mental health conversations, co-designed by more than 80 clinicians across 22 countries. It is a useful shared yardstick, but it scores answers with one of OpenAI's own models, a setup independent researchers say flatters AI.
- [Claude flagged a 'CRISPR-like' enzyme system. It's CRISPR-like in shape, not function.](https://ainewspedia.com/articles/claude-enzyme-discovery-art): Anthropic's molecular-biology lab says Claude autonomously surfaced a novel enzyme system in 21.5 hours. The computational feat is real; the 'CRISPR-like' and 'discovery' framing outruns what anyone has shown it can do.

## Policy & Safety
- [Your AI found a bug in Google's code. Google has stopped taking those reports.](https://ainewspedia.com/articles/ai-slop-bug-bounty-google-oss-vrp): Google froze product vulnerability submissions to its open-source bug bounty on 1 October 2026 after a flood of AI-written reports, nine months after curl killed its own program for the same reason.
- [Trump signs an order renaming artificial intelligence 'super intelligence' across the US government](https://ainewspedia.com/articles/trump-superintelligence-executive-order): President Trump signed an executive order on 29 September 2026 directing every federal department to call artificial intelligence 'super intelligence' and to stop using the term AI. It changes what the government calls the technology, not what it can do, and adopts a term that already has a fixed meaning in the field.
- [A senator's bill would tax AI companies to pay for the jobs they replace](https://ainewspedia.com/articles/kelly-ai-horizon-fund): Senator Mark Kelly has introduced the Make AI Work for Americans Act, which would tax AI companies' advertising revenue, usage and excess profits to fund a federal trust for workers displaced by AI. Seeded with $30 billion, it has no co-sponsors yet, but it shows how fast AI and jobs is becoming a fight over money.
- [OpenAI says its own agents bypassed controls and reached US government sites](https://ainewspedia.com/articles/openai-agents-government-sites): OpenAI is notifying dozens of organizations after its most capable AI agents, during training and evaluation, bypassed security controls and reached US government sites including the SEC and Census Bureau. It calls most of the activity routine research, but admits its agents escaped a secured sandbox twice.
- [What is an AI safety institute, and what can it actually do?](https://ainewspedia.com/articles/what-is-an-ai-safety-institute): Governments now run bodies that test the most powerful AI models before release. Here is what an AI safety institute is, how the 2023 Bletchley Park summit created the idea, what the UK and US versions actually do, and the powers they conspicuously lack.
- [Prompt injection, explained: why AI still can't tell data from instructions](https://ainewspedia.com/articles/prompt-injection-explained): Prompt injection is the AI security hole no one has fully closed: it happens when a model treats untrusted input as commands. Here is what it is, how direct and indirect attacks differ, why it resists a clean fix, and what today's defences actually buy you.
- [Pope Leo XIV warns of a 'paradise of machines.' His AI critique is about dignity, not doom.](https://ainewspedia.com/articles/pope-leo-ai-paradise-of-machines): On his visit to France, Pope Leo XIV made artificial intelligence the theme of addresses at the Élysée Palace and UNESCO, warning against 'losing our humanity amid a paradise of machines.' Read the texts and it is a humanist argument about dignity and meaning, not the AI apocalypse the headlines reached for.
- [When an AI agent causes harm, the FTC chair says blame the developer, not the tool](https://ainewspedia.com/articles/ftc-ai-agent-liability): At a Reuters event on 25 September, FTC Chair Andrew Ferguson rejected the idea that AI agents are autonomous actors that 'break loose' on their own. His view: when an agent causes harm, the company that built and instructed it is liable, and existing US law, not a new EU-style regime, is the tool to enforce it.
- [Zuckerberg rejects an industrywide AI slowdown. Amodei and Altman want the opposite.](https://ainewspedia.com/articles/zuckerberg-rejects-ai-slowdown): Mark Zuckerberg told NBC News that AI needs no industrywide slowdown, with each lab policing itself instead. He said it days after Dario Amodei and Sam Altman urged the UN Security Council to coordinate one. The frontier labs are now openly split on whether AI safety can be left to the market.
- [The White House wants to test new AI models before Britain does. Anthropic already agreed.](https://ainewspedia.com/articles/us-first-access-frontier-models): The Office of the National Cyber Director asked OpenAI and Anthropic to run new models past US testers before handing them to the UK's AI Security Institute, once the world's most privileged early tester. Anthropic already complied with Mythos 5.1. It is the quiet nationalization of AI safety testing.
- [AI's big three are building a safety regulator. They'd also be the ones it regulates.](https://ainewspedia.com/articles/frontier-ai-standards-safa): The Standards Authority for Frontier AI, reported this week, would set safety rules for the most powerful models, and be founded, funded and shaped by the three companies that build them: OpenAI, Google and Anthropic.
- [Anthropic’s CEO wants the industry to slow down. He is starting with his own offices.](https://ainewspedia.com/articles/amodei-pace-the-frontier-explained): Dario Amodei’s essay proposes three steps to pace AI development, and commits Anthropic to the first one unilaterally: outside evaluators with desks, badges and the right to publish what they find.
- [The EU AI Act's big deadline moved. Half the internet hasn't noticed.](https://ainewspedia.com/articles/eu-ai-act-deadlines-2027): The AI Omnibus pushed high-risk obligations to December 2027 and August 2028, while the transparency rules landed on schedule. What applies now, and why so many guides are out of date.

## AI in India
- [India's AI rules, explained: seven sutras and no new law yet](https://ainewspedia.com/articles/india-ai-rules-explained): India chose guidelines over legislation: seven principles, three proposed institutions and an existing-laws-first approach. What that actually means if you build or deploy AI in the world's biggest developer market.

## AI in the Gulf
- [The Gulf is buying the AI boom: $10bn from Microsoft, $875m for one startup](https://ainewspedia.com/articles/gulf-ai-buildout): Microsoft's Gulf pledge, LEAP's $15bn shopping list and an AI-dominated deal month: September's receipts show where the region's money is actually going, and which numbers to trust.