Presented by
Good morning, AI enthusiasts. The U.S. Treasury Secretary and the head of the Federal Reserve just called America's biggest banks to warn them about a single AI model.
If you've been following the Anthropic Mythos story, this is the week it stopped being a tech story. When Bessent and Powell pick up the phone together, the calculus changes.
In today's recap:
Bessent and Powell summon Wall Street over Mythos risks
Perplexity adds Plaid-powered personal finance hub
Fact-check AI outputs with Perplexity
Google AI Overviews serving false answers by the minute
4 new AI tools, prompts, and more
WALL STREET
Bessent and Powell summon bank CEOs over AI risks
Kent Nishimura/Bloomberg
Recaply: Treasury Secretary Scott Bessent and Fed Chair Jerome Powell just convened an emergency meeting with Wall Street bank executives to warn them about cyber risks from Anthropic's Mythos model, marking the first time top U.S. financial regulators have mobilized over an AI product.
Key details:
The Treasury Department meeting aimed to ensure major banks know about Mythos risks and are taking active steps to harden their systems against the model's advanced hacking capabilities.
Bessent and Powell's joint call is the first time the two highest U.S. financial regulators have met with bank CEOs over an AI threat, with Anthropic's run-rate revenue now at $30B, up from $9B at the end of 2025.
Bloomberg News first reported the meeting, citing sources familiar with the talks. Reuters could not immediately verify the report.
The meeting took place Tuesday at the Treasury Department, following Anthropic's limited Mythos Preview rollout earlier this week to a select group of cybersecurity partners.
Why it matters: Regulators have watched AI for years. But something changed this week. The Treasury Secretary and the Fed Chair, two officials who usually deal with bank runs and market crashes, just ran emergency briefings over one AI model. Mythos isn't just a cybersecurity product anymore. The U.S. government is treating it the same way it once treated Lehman Brothers, as a potential threat to the financial system itself.
PRESENTED BY MINTLIFY
AI agents now read your docs almost as much as humans do.
Mintlify analyzed 790 million requests across its documentation platform. The finding: AI coding agents account for 45.3% of all traffic, nearly tied with traditional browsers at 45.8%.
Two tools are driving almost all of it:
Claude Code: 25.2% of total traffic, more requests than Chrome on Windows
Cursor: 18% of total traffic
Together they account for 95.6% of all identified AI agent traffic
The rest of the field, OpenCode, Trae, ChatGPT, and NotebookLM, is showing up but nowhere close.
One caveat: OpenAI's Codex doesn't send an identifiable user-agent header, so the real agent percentage is likely even higher.
The takeaway for anyone maintaining developer docs: your documentation now serves two audiences. Structure and machine-readability matter as much as clarity for human readers.
PERPLEXITY
Perplexity turns AI search into a personal finance hub
Perplexity
Recaply: Perplexity just launched a full personal finance feature through Plaid, letting users link bank accounts, credit cards, and loans to the AI tool for live spending data and custom money dashboards.
Key details:
Once accounts are linked, Perplexity can track spending by type, show net worth across all accounts, and answer free-form money questions without preset menus.
Plaid works with 12,000+ banks, including Robinhood, Chase, Vanguard, and Fidelity. More than 75% of Perplexity users visit monthly to ask financial questions.
The integration uses read-only access and user data never touches Perplexity's servers. Live data from FactSet, Coinbase, Nasdaq, and S&P Global powers the analysis.
The feature launched April 9 for signed-in US and Canada desktop users. Custom tools like budget trackers and savings dashboards are only for Pro and Max plan subscribers.
Why it matters: Most budget apps give you dashboards someone else built. Perplexity lets you ask "build a debt payoff plan using my credit card balances" and it figures out the structure. With 75% of its users already coming for money questions, the Plaid integration doesn't add a new use case. It turns a workaround people already had into an actual product.
GUIDES
Fact-Check AI content fast with Perplexity
Recaply: In this tutorial, you will learn how to use Perplexity AI to verify claims in AI-generated content, so your team stops sending out false information by accident.
Step-by-step:
Copy the AI text and mark every claim that needs checking: stats, dates, product names, quotes, and news items. These are your targets for the session.
Open Perplexity and enable Pro Search using the toggle above the search bar. Paste one claim at a time as a direct question: "Is this accurate? Find me sources for: [claim]."
For each answer, hover over cited figures to see the original source link. Click through to confirm the source actually backs up what Perplexity says it does.
If Perplexity returns no supporting sources for a claim, or flags the answer as ungrounded, remove the claim or rewrite it as an open question rather than a fact.
Build a Perplexity Space just for fact-checking with a saved prompt template. That way, your team can run the same process on any AI-generated piece without starting from scratch each time.
Pro tip: For claims about earnings or product launches, set the date filter to "Past month" to make sure Perplexity pulls from current sources, not outdated cached pages.
TOOLS
Trending AI Tools
🛡️ Bouncer - Free X feed filter that removes content you don't want based on a plain text description
🤝 Gooseworks - AI coworkers with their own computer, email, and memory for GTM work
📱 Shipper - Turns any website URL into a native iOS/Android mobile app
🎥 Tinycloud - AI agent for video work
NEWS
What Matters in AI Right Now?
Startup Oumi found that Google's AI Overviews are inaccurate 9-15% of the time, with 51% of Gemini 3 answers "ungrounded" by the cited sources. With 5 trillion searches expected in 2026, that works out to hundreds of thousands of false answers every minute. Google disputed the methodology.
Anthropic is exploring designing its own AI chips to reduce reliance on Nvidia, according to three sources. Plans are early stage and no dedicated team has been formed, though advanced AI chip design can cost roughly $500M.
A three-judge D.C. Circuit panel rejected Anthropic's request to pause the Pentagon's supply chain risk label, which blocks its products from DOD contracts. The label, never before applied to a U.S. company, stems from Anthropic's refusal to let the Pentagon use Claude for autonomous weapons or mass surveillance.
OpenAI is finalizing a new product with advanced cybersecurity capabilities for release to a small group of partners first, mirroring Anthropic's limited Mythos rollout. Both companies are now restricting model access due to fears about the hacking potential of their frontier models.
OpenAI introduced a new $100/month Pro tier for ChatGPT with 5x more Codex usage than the Plus plan, built for longer and more demanding coding sessions. Through May 31, new Pro subscribers get up to 10x Codex usage compared to Plus.
LM Studio acquired Locally AI, bringing its creator Adrien Grondin onto the team to lead native AI experiences across Apple devices. Locally AI is a popular iPhone, iPad, and Mac app for running AI models on device.
Intel and Google announced a multiyear deal to advance AI and cloud infrastructure, with Intel Xeon processors continuing to power Google Cloud and both companies expanding co-development of custom IPU chips. Intel CEO Lip-Bu Tan said scaling AI "requires more than accelerators."
Google Research introduced ConvApparel, a new human-AI conversation dataset for measuring the "realism gap" in LLM-based user simulators, published at EACL 2026. The dataset uses a dual-agent protocol with both helpful and intentionally unhelpful AI agents.
CoreWeave announced a $21B expanded AI cloud deal with Meta through December 2032, covering multiple locations and including early deployments of the NVIDIA Vera Rubin platform.
EVENTS
Google Cloud Next 2026: April 22-24, 2026 • Las Vegas, NV
Anthropic Code with Claude: May 6, 2026 • San Francisco, CA
Google I/O 2026: May 19-20, 2026 • Mountain View, CA / Online
NVIDIA GTC Taipei 2026: June 1-4, 2026 • Taipei, Taiwan
Microsoft Build 2026: June 2-3, 2026 • San Francisco, CA / Online
🧡 Enjoyed this issue?
🤝 Recommend our newsletter or leave a feedback.
How'd you like today's newsletter?
Cheers, Jason
