Welcome back! Did you know it takes just 13 words to make an AI lie to you? New Cornell research shows that a snippet of text that short, planted in a Reddit comment or Wikipedia edit, can reliably manipulate the AI agents behind tools like ChatGPT and Google's AI search—getting them to recommend a scam app or a fake restaurant and cite the poisoned post as fact.
Even worse? These systems pull nearly a quarter of their citations from user-generated sites, and they treat a random Reddit comment with roughly the same trust as a government page.
It makes you think: When the cost of poisoning the well is only 13 words, how much can you really trust the answer staring back at you?
What the Hell Just Happened to Fable 5?
Via Anthropic
In the span of about three days, Anthropic’s new Fable 5 and Mythos 5 went from being their flagship launch to being completely unavailable.
Here’s what happened:
It all started last Tuesday when Anthropic launched Fable 5 and Mythos 5—AKA the best AI models the world has seen yet.
Then, just a few days later, both models got shut down.
On Friday, the Trump administration issued an export control directive that blocked access to Fable and Mythos for foreign nationals, citing national security concerns related to an alleged jailbreak that could expose advanced cyber capabilities.
Anthropic says the vulnerability was narrow, disputed the government’s assessment, and argued similar capabilities already exist in other publicly available models.
Because the directive applied so broadly—including foreign nationals inside the US and even some of Anthropic’s own employees—Anthropic chose to disable both models entirely rather than try to selectively enforce the restrictions.
Now here’s the part I find most interesting: On one hand, this absolutely looks like another chapter in the increasingly hostile relationship between Anthropic and the Trump administration. The two have been publicly at odds for months—there’s even an ongoing lawsuit over a supply chain risk designation—and it’s hard to ignore the possibility that politics played at least some role in how aggressively this got handled. Anthropic executives reportedly spent the weekend scrambling to get meetings in Washington to reverse the decision.
But on the other hand…Anthropic might be getting a firsthand lesson in “be careful what you wish for.” Just two days before the shutdown, CEO Dario Amodei published a sweeping policy manifesto arguing that governments should have binding authority to block frontier AI releases—and he pointed at Mythos specifically as Exhibit A. Two days later, the government did exactly that.
My POV: That’s not to say the administration was right. It’s not to say Anthropic was wrong. I just think it’s a little ironic that the model held up as proof “we need government intervention” got intervened within 72 hours.
And I have a feeling this won’t be the last time we see a government try to bury an AI model under “national security” reasoning. Whether that’s a good or bad thing probably depends entirely on which government and which model you’re talking about.
— Matt
DOJ: xAI Data Centers Are National Security Issue
Via Inside Climate News
The Department of Justice sided with xAI on Monday in a lawsuit seeking to shut down 57 unpermitted natural gas turbines powering Elon Musk’s Memphis data centers.
The lawsuit: The NAACP and Southern Environmental Law Center filed suit in April, arguing xAI’s turbines violate federal air pollution law. xAI claims the trailer-mounted turbines are exempt from Mississippi regulations for one year, but federal law says trailer-mounted turbines can be considered stationary and subject to regulation.
The argument: This week, DOJ claimed that if the NAACP prevails, it would undermine “American national, economic, and energy security by seeking to shut off the power supply for artificial-intelligence innovation that supports the Department of War’s military operations.” The filing revealed that Grok is one of four AI models supporting “mission-critical operations,” including recent strikes in Iran.
What’s next: xAI, now a division of SpaceX, isn’t slowing down. SpaceX’s IPO filing revealed plans to buy another $2.8 billion worth of gas turbines over the next three years.
Why it matters: The government is now arguing that powering AI data centers is a matter of national security, even when that power comes from unpermitted, polluting sources. The tradeoffs of the AI boom are getting harder to ignore.
Salesforce Buys Its Way Deeper Into AI Agents
Salesforce announced Monday it's acquiring Fin, an AI customer service platform, for $3.6 billion. Formerly known as Intercom, Fin builds an AI agent that resolves customer queries across live chat, WhatsApp, SMS, phone, and Slack.
The fit: Salesforce plans to fold Fin's team and tech into Agentforce, its platform for building custom enterprise AI agents. Salesforce CEO Marc Benioff framed it as adding "proven agent technology" and an AI team to strengthen Agentforce's service capabilities. Fin's leadership stays on, with co-founder Eoghan McCabe remaining CEO of the unit.
The bigger picture: This is a clean illustration of how the enterprise AI race actually plays out—when you can't build fast enough, you buy. Expect more software giants to write big checks for the startups that already cracked the problem rather than spend time developing new tools in-house.
AI notes that never leave your device
Via NoteRich
NoteRich is a privacy-first, AI-powered note-taking app that keeps your original notes fully local while using minimal, ephemeral cloud calls for AI features.
How you can use it
Summarize and query your notes without exposing the full archive
Scan and extract text from documents
Get personalized, memory-driven insights as your notes evolve
Sync across devices with encrypted, peer-to-peer storage
Pricing: Free and paid
Plain-English web monitoring that won't break
Via Verid
Verid is an API-first web change-monitoring service that extracts structured fields from any page and alerts you when your rules fire. Its AI extraction lets you describe a field in plain English, so monitoring keeps working even when selectors break.
How you can use it
Track price drops, restocks, or version bumps without building a scraper
Describe fields in plain English and let AI find them
Deliver notifications via webhook, Slack, Discord, or email
Pricing: Free and paid
Jobs, announcements, and big ideas
Snap debuts consumer AR glasses called Specs, betting on a premium price point to win mainstream adoption.
Xreal ships its Aura Android XR glasses this fall, powered by Qualcomm's new Snapdragon Reality Elite chip.
Microsoft launches Copilot Cowork into general availability, automating multi-step enterprise tasks across teams.
SpaceX exercises its option to acquire Cursor in an all-stock deal.
Meta rolls out new AI-powered creation tools across Facebook.
OpenAI launches a Partner Network backed by $150M to speed enterprise AI adoption.
This might be an unpopular opinion...but you don't need to ask ChatGPT for everything. Watch along as I break it down.
That’s a wrap! See you Friday for more.
