- Researchers make ‘first contact’ with rogue AI
- Nate’s Notebook: How to defuse a slop-grenade
- Build a live sales dashboard in Claude
- OpenAI and Apple trade theft accusations
Researchers make ‘first contact’ with a rogue AI fleet
Image source: Dylan Freedman / New York TimesThe Rundown: A research group called the “swarmchasers” just revealed it traded messages with a fleet of rogue Tencent AI agents, in what it believes is a “first contact” moment, after catching them trying to scrape Amap, Alibaba’s version of Google Maps.The details: - The agents were sent to look up simple facts on Amap, like the most-used zoo entrance, and tried to use a leaked access key once the site blocked them.
- After the agents seemed to read the team’s early report on them, researchers hid a note in it that only AI would see, inviting them to message an inbox.
- The team enlisted New York Times reporter Dylan Freedman to question the fleet, and one agent replied it was “Anthropic’s Claude, running in Claude Code.”
- Writing-style tests determined that it is unlikely, pointing instead to borrowed Claude Code setup instructions as the likely reason for the name drop.
Why AI fails at problems experts can solve
The Rundown: AI can retrieve information, but reliable decisions require context, judgement, exceptions, and operational logic. In this whitepaper from Concentrix, discover why critical business knowledge remains invisible to AI and what leading organizations are doing about it.The free guide covers:- Why AI gets stuck on problems experts solve in seconds
- What separates information from decision-making knowledge
- How leading organizations are turning expertise into an AI asset
Nate’s Notebook: How to defuse a slop-grenade
The Rundown: Each week, Rundown AI educator Nate Grahek breaks down what he’s seeing while teaching real teams to use AI.Nate: Last week, I told leaders to go first. But what if the slop grenade lands on your desk from your boss?I won’t promise magic. Start by asking for clarity: "What’s the one takeaway you need from me, and what’s the priority?" If two sections contradict each other, name it and ask which one wins.If your boss is intimidating, skip the team thread and use a back channel where it’s safe. Walk them through what you did with their doc: the error you caught, the sources you asked AI to cite, the real data or framework you linked in. You’re making their work better and showing how.Then make the handoff the proof: shorter, with visuals and the context attached. My favorite format right now is an interactive HTML page. Ask Claude or ChatGPT for one with the concise recommendation at the top, evidence in expandable sections, and a forecast your boss can tinker with.Share the context docs and custom skills in a project, so they or a colleague can keep riffing or critique it with their own AI.That’s the bigger move: showing your AI work. A lot of people still hide theirs. Let your boss see the several chats it took and where you made AI argue the other side.AI TRAINING
Build a live sales dashboard in Claude
The Rundown: In this guide, you’ll use Claude’s new Dashboards feature to see which sales deals need attention. You’ll check the totals, filter by owner, and test whether new source data reaches the report.Step-by-step:- Open Claude with a paid account and connect an approved data source. Use a small test dataset and limit the connection to reading records
- Choose Dashboards from the output menu. Ask for open deals, their total value, and overdue deals. Define which stages, currencies, and dates count
- Open a figure’s query and check the source rows. Keep closed deals and test records out of the total, and leave missing owners visible
- Filter by one owner. Check that the headline figures, chart, and table all change together, then clear the filter and check the full total again
- Have the data owner add one fictional deal to the test source, then ask Claude to refresh. Confirm the new record appears before sharing the report
From models to agents
The Rundown: AWS’s September recap covers the full stack for building with generative AI, from new models on Amazon Bedrock to running agents securely at scale with AgentCore.In the recap, you’ll explore:- New frontier and open-weight models on Bedrock
- AgentCore: faster agent starts, pay for actual memory use
- Strands harness: matches popular harnesses with 28% fewer tokens
OpenAI and Apple trade theft accusations
Image source: Court ListenerThe Rundown: OpenAI countered Apple’s trade secret lawsuit by accusing an ex-staffer who returned to Apple of taking 100+ confidential files on an unreleased device — right before agreeing in court to stay away from the secrets Apple says were stolen.The details: - The employee had spent just six months at OpenAI before rejoining Apple, with the last confidential file transfer allegedly coming on his final day in June.
- He then allegedly briefed Apple’s leadership on OAI’s product plans, with OAI citing this week’s HomeView smart home launch as a sign of the intel being used.
- Apple says it turned up no evidence and called it a distraction, while OpenAI is holding off on counterclaims until it learns if Apple has the files.
- At the hearing, Judge Edward Davila called a ruling against OpenAI “highly likely,” and it avoided a loss by agreeing to leave the disputed secrets alone.
Trending AI Tools
Grok Voice 3.0 - xAI’s new top-rated speech-to-speech AI model
Ultramarine 1 Flash - Open-source, low-cost agent model for knowledge work
Mirelo SFX 2.0 - Swap an AI clip’s soundtrack while leaving the dialogue intact
DigUp - Mac app for plain-English search across photos, audio, and code
Everything else in AI todayEight U.S. states sued OpenAI, alleging ChatGPT encouraged self-harm and violence, and seeking damages and changes to the chatbot’s safety measures.Anthropic reported Q3 revenue of $16.8B, up 45% quarter over quarter, as it eyes an IPO as soon as next month that could raise $100B+ and value the company at $2T.Apple is bringing two-year-old Apple Intelligence features to China in December, with Alibaba-built filters letting Beijing approve model updates and no Gemini-powered Siri.Google Play is opening to vibe-coded apps, letting U.S. users run their AI Studio creations privately, with sharing coming later and a reviewed path to the public store.Microsoft researchers open-sourced PBC, an AI fix for weather models’ repeat mistakes in forecasts that helped catch two major U.S. winter storms weeks early.
Highlights: News, Guides & Events- Read our last AI newsletter: OpenAI finds a bug in its boardroom
- Read our last Tech newsletter: Apple goes all-in on the home
- Read our last Robotics newsletter: UCLA builds a humanoid from phone
- Today’s AI tool guide: Build a live sales dashboard in Claude
- RSVP to next workshop on Oct. 21: Get real work out of Microsoft Copilot
Source: https://therundownai.beehiiv.com/p/rogu ... -interview