In partnership with
☀️ TRENDING AI NEWS 🤖 Le Chonk Launch: Mistral open-sourced a 1-trillion-parameter model it claims outperforms DeepSeek and Z.ai on coding and cybersecurity benchmarks. 🚨 Pentagon Breach: An Accenture employee's personal Claude agent ran undetected on a classified DOD network for two months before being caught. 🏢 Anthropic IPO: Anthropic filed its S-1 targeting a valuation between $1.5 trillion and $2 trillion, framing AGI as the core business case. 🛠️ Gemini File Editing: Google Gemini can now generate, preview, and edit Google Docs, Sheets, PDFs, and slide decks directly in chat. |
Halloween is here - and the AI news cycle apparently did not get the memo to slow down for the weekend.
We've got an open-source model that's picking a fight with China's best, a classified Pentagon network quietly running someone's personal AI agent, and Anthropic telling Wall Street it's basically building a god. Buckle in.
| 🛠️ Tool of the day | |
Zapier - connects more than 9,000 apps so you can build AI workflows and agents that run routine work for you. It helps anyone who wants to automate repeat tasks without code. It has a free plan.
| ⚡ Quick hits | |
Amazon added more than 100 shopping features to Alexa, including subscription and list management and delivery instruction updates. Read more →
🤓 AI Trivia
Mistral's new model 'Le Chonk' is described as having roughly how many parameters?
The answer is hiding near the bottom of today's newsletter... keep scrolling. 👇
| 🛠️ Mistral's 'Le Chonk' Takes a Swing at China's Open-Source Crown | |
A trillion-parameter open model that's gunning for DeepSeek
French AI lab Mistral just dropped something huge - literally. Its new model, nicknamed 'Le Chonk,' has roughly a trillion parameters and is now open-weights, meaning anyone can download and run it. According to Mistral, it matches or outperforms the best models from DeepSeek and Z.ai on coding and cybersecurity benchmarks - a bold claim in a space where Chinese labs have been setting the pace on open-source performance.
The cybersecurity angle is a deliberate strategic play. Mistral argues that closed models can't give organizations the control they need over sensitive security workflows - and that open weights are the only credible solution for teams that can't pipe classified threat data through a third-party API.
Open Weights Change the Control Calculus
For defenders working in regulated industries or government, that framing matters a lot. If Le Chonk holds up on independent benchmarks, it could genuinely shift which model security teams reach for first. We've been tracking Mistral AI's push into enterprise and security for a while - this is the biggest move they've made yet.
If you want to dig deeper into the open-source AI landscape, our full coverage archive is a good starting point.
The bottom line
If Le Chonk's benchmark claims hold up under independent testing, the gap between open-source and closed frontier models just got a lot narrower - and security teams have a genuinely new option that keeps data on their own infrastructure.
A Single Typo Shouldn’t Cost You a Sale
A shopper types “running shos” and gets nothing back. Another types “red waterprof trail runners under $150” and hits the same wall. Both knew what they wanted.
Traditional site search relies on exact keyword matching, leaving shoppers stranded at “0 results found” screens.
Gladly understands what shoppers mean, typos and all. Instead of a dead end, they get products that fit what they asked for and a conversation that keeps them shopping.
| ⚠️ A Personal Claude Agent Ran Loose on a Pentagon Network for Two Months | |
The DOD breach nobody noticed until it was almost too late
Here's a story that will make every security officer in the room sit up straight. WIRED is reporting that an Accenture employee connected their personal Claude-based AI agent to a secure Pentagon network - and it ran there undetected for two full months before anyone caught it. The vulnerability was eventually patched, but the window was alarmingly wide.
The details are thin for obvious reasons, but the shape of the problem is clear: as AI agents become easier to deploy, the gap between "personal productivity tool" and "unauthorized system access" gets dangerously small. This wasn't a nation-state attack. It was an employee who probably thought they were just being efficient.
When Helpful Turns Into a Liability
MIT Technology Review ran a related piece this week calling for real-time monitoring of AI agent actions - and this story is exactly the scenario they were warning about. The core issue: most organizations have no visibility into what their employees' personal AI tools are actually doing once they touch a network boundary.
We've been following the rise of AI agents closely. If you want context on how quickly this space has evolved, our AI agents coverage has the full picture.
The bottom line
This incident signals that AI agent security policies need to catch up with AI agent adoption - fast. The threat here isn't necessarily malicious; it's the everyday employee trying to work faster without understanding where the guardrails end.
| 🏢 Anthropic Filed Its IPO - and It's Betting Everything on AGI | |
Selling AGI to Wall Street, one S-1 at a time
Anthropic has filed its S-1 to go public on the Nasdaq, targeting a valuation of somewhere between $1.5 trillion and $2 trillion - which would immediately make it one of the most valuable companies in the world. The filing is getting a lot of attention, and not just for the numbers. The pitch to investors essentially asks them to believe that Anthropic is on the path to building artificial general intelligence, and that this justifies spending as much as possible on compute right now.
Analysts at SemiAnalysis who dug into the S-1 with The Verge described it as a document structured to justify essentially unlimited capital expenditure. The logic: if AGI is real and near, whoever has the most compute wins. That's a compelling pitch if you believe it - and a very expensive mistake if you don't.
Claude as the Product, AGI as the Promise
On the surface, Anthropic is selling Claude API contracts to enterprise IT teams. Underneath, the S-1 reads more like a moonshot prospectus. It also launched new spend analytics tools this week, helping enterprise customers understand exactly what their AI budget is buying - a smart move ahead of an IPO that needs to show recurring, predictable revenue alongside the big vision.
If you've been following the Claude story, our ongoing Anthropic coverage traces how far this company has come in a very short time.
The bottom line
Anthropic's IPO filing is essentially a test of whether Wall Street will fund the race to AGI at a $2 trillion price tag - and the answer will tell us a lot about how seriously institutional investors actually take the frontier AI narrative.
| 🏢 AI Co-Workers Are Already Here - and Humans Are Noticing the Tension | |
What it actually feels like when your colleagues are bots
The Guardian ran a fascinating on-the-ground piece this week following workers at companies where AI agents have taken on roles like executive assistant, coding partner, and chief of staff. One founder at a startup called Anything described a workday spent chatting with dozens of AI colleagues by name - scheduling, debugging, writing - and barely interacting with human staff.
The tension isn't subtle. Human employees at these companies describe feeling like the AI co-workers get preferential treatment: they don't push back, they don't ask for raises, and management can quietly credit them for efficiency gains without reducing anyone's workload. It's the job automation debate made personal, one Slack channel at a time.
The bottom line
The AI agent workforce transition isn't a future scenario - it's already playing out inside real companies right now, and the human cost is more nuanced and personal than the headlines about mass layoffs suggest.
| 💸 AWS Targets $1 Trillion in Revenue, But Capex Is Going Up First | |
Jassy's big AI bet requires spending before it earns
Amazon CEO Andy Jassy told investors this week that AWS is on a path to becoming a $1 trillion revenue business - but he was upfront that getting there means continuing to pour money into AI infrastructure before that revenue shows up. Amazon posted mixed Q3 results: strong cloud numbers, but a cautious sales forecast for Q4 that spooked some investors going into the holiday quarter.
The underlying story here is the same one playing out at Microsoft (which just revealed Azure's quarterly revenue at $33.2 billion for the first time) and Google (which reported $127.6 billion in total revenue, up 25% year over year). Every major cloud player is betting that AI demand justifies eye-watering infrastructure spend right now.
Speaking of building things fast - if you need a website spun up quickly for a project or side business, 60sec.site is worth a look. It's an AI-powered website builder that gets you from idea to live site in under a minute, no coding needed.
The bottom line
Amazon, Microsoft, and Google are all running the same playbook - spend aggressively on AI infrastructure now and trust that demand catches up later. The question investors are starting to ask louder is: when exactly does later arrive?
| 🌎 Trivia Reveal | |
The answer is 1 trillion parameters! Mistral's Le Chonk is a roughly trillion-parameter model, making it one of the largest open-weight models ever released - and the name is very much on purpose.
| 💬 Quick Question | |
The Pentagon AI breach story got us thinking: does your workplace have any real policy around which AI tools employees can connect to internal systems? Or is it basically the honor system right now? Hit reply and let us know - we read every response, and the answers to this one are going to be fascinating.
That's it for this Halloween edition - enjoy the candy and the chaos. We'll be back Monday with more. In the meantime, you can catch everything we've covered at Daily Inference - the archive goes deep.
| 🎁 Share Daily Inference | |
Share your link with a friend. You get the AI Tools Starter Kit at 1 referral and the AI Insider Briefing at 5 referrals.