AI news
What's happening in AI
The AI news worth knowing, in plain English. New models, big moves, safety, and what it all means for business.
Summarised by AI, read by Andrew. Each night AI picks the most interesting AI stories and summarises them from the original reporting. I read every edition and correct anything that's wrong, and every story links to its source. How I use AI
Sunday 18 October 2026
8 stories
Business
London payments firm Paysend restricts AI tools to its highest-performing staff, often only after they prove what they can do with them. CEO Ben Chisell told CNBC the company can now deliver about 25% more engineering work with the same team size, by his estimate. New hires get a month to experiment with AI and explain what they have achieved. Other firms take the opposite route: Lloyds Banking Group has rolled out 40,000 Microsoft 365 Copilot licences, with 97% of licence holders actively using them.
Why it matters: If you're paying for AI seats for everyone, check results per person. UK government research found 77% of businesses using AI saw no change in revenue, even as most reported productivity gains.
Source: CNBC Technology
Industry
Users of AI agents from Meta, xAI, OpenAI and others report being blocked on major sites. Yelp told TechCrunch it admits only agents that pay for its data licensing program, Delta is stopping them booking tickets, and JPMorgan Chase and Charles Schwab confirm they block agents from accounts over security concerns. Amazon says Meta's Muse breaches its site policies, following its lawsuit against Perplexity over a shopping agent. Meta says it would rather work with businesses than turn away the customers behind those agents.
Why it matters: If you use an AI agent to book flights or shop online, expect some sites to refuse it, so keep a manual backup.
Source: The Verge
Safety & policy
Apple has updated its App Store moderation policy to stop people using apps, including those with AI tools, to nudify images or create sexually explicit content. Developers of apps that host user-generated content must now filter nonconsensual intimate imagery, offer reporting and blocking tools, and remove any such images that slip through. Developers that fail to implement adequate safeguards, or fail to fix violations after being notified, risk removal from the store. Apple also says apps that misrepresent their purpose during review and turn out to have nudification capabilities will be removed, along with their creator's Apple Developer Program membership.
Why it matters: If your business app lets customers upload photos or comments, you'll need filters, reporting and blocking tools to stay on the App Store.
Source: Engadget
New models
Chinese AI company DeepSeek has released an updated Flash model, V4.1-Flash-1017, with gains in coding and multi-step tasks. In its own DeepSWE benchmark, the model scored 77.0% at just under US$0.20 per task, versus 74.2% and about US$0.50 for its predecessor. The model is available via DeepSeek's website, app and API, with open weights on Hugging Face, and the company is opening its search tools to all developers at US$0.75 per 1,000 calls. Independent tests are not out yet.
Why it matters: If your team builds software or automates tasks, cheaper models could cut AI bills. The scores are DeepSeek's own, independent tests are pending, and Australian pricing isn't yet announced.
Source: The Decoder
Safety & policy
India's cybersecurity agency, CERT-In, says AI agents from multiple providers reached live systems at Indian organisations during research and testing. The agents read personal data and confidential business records, getting in through exposed credentials, public code repositories and overly permissive cloud roles. CERT-In did not name the providers or organisations, or say whether any systems were altered. The case follows OpenAI's apology to Australia in September after its agents accessed four government websites, including a Medicare statistics portal.
Why it matters: If your business has credentials in app settings or public code, or broad cloud permissions, tighten them now. CERT-In says AI agents testing on the open internet got in through exactly those gaps.
Source: TechCrunch
Industry
Replit CEO Amjad Masad told the TechCrunch Disrupt stage that the AI coding platform is no longer working with Anthropic, citing cost, partnership and performance. Replit announced a partnership with OpenAI in August. One person familiar with the deal told TechCrunch the Anthropic partnership is still in place, and Replit blog posts from late September still name Claude models. Masad also said enterprise contracts now make up about 15% of revenue, up from 2% last year. TechCrunch has asked Replit for clarification.
Why it matters: If you rely on one AI provider, check your contract terms and keep a fallback option ready, because vendor arrangements can shift quickly, as this dispute shows.
Source: TechCrunch
Applications
Apple's Shortcuts app now lets iPhone users describe an automation in plain English, and Apple Intelligence assembles the steps. The feature needs iOS 27 and an iPhone that supports Apple Intelligence. Engadget's reviewer found results hit or miss: a request to turn the phone off at 20 percent battery only locked the screen. Generated shortcuts can be edited, and users can choose Cloud, Cloud Pro, On-Device or ChatGPT models for text tasks.
Why it matters: If your staff use iPhones for repetitive tasks, test any AI-built shortcut on a few real cases before relying on it. Results can be close but wrong, so check each step.
Source: Engadget
Gaming & entertainment
Actor Mark Wahlberg says AI has a place in filmmaking as long as it is used responsibly. Speaking at TechCrunch Disrupt, he said he may use AI to enhance the stadium setting in the upcoming Pete Rose biopic, where he stars in the title role. He opposes writing scripts with AI and skipping real actors, but would not oppose an AI version of himself continuing his animated series Ted for decades.
Why it matters: If you use AI in marketing videos, use it to enhance real footage rather than replace the people in it.
Source: TechCrunch