Welcome, humans.

ANNOUNCEMENT: We're hiring a Newsletter Sales Account Executive!

Think you can convince brands that our AI-obsessed readers are the perfect audience for their products? We need someone to sell ad space in The Neuron (and our other newsletters).

This isn't your typical “spray and pray” sales gig. You'll actually get to work with engaged audiences who want to discover new tools. Plus, you'll help build something from the ground up as we expand our newsletter empire.

If you've got 2+ years of media advertising sales experience and can close deals without being pushy, we want to chat. Remote-first, solid benefits, and the chance to shape our ad strategy from day one.

Apply here before someone else beats you to it!

Here’s what you need to know about AI today:

• We recap the top 5 WILDEST stories from OpenAI’s latest threat report.

• Gemini 2.5 Pro is now the top AI model, barely overtaking ChatGPT o3.

• Google’s “free” AI Studio tool is set to become “API based” (paid).

• Apple released a paper that shows “thinking AI” have serious limits.

OpenAI's New Threat Report is Full of Spies, Scammers, and Spammers… Here’s the Top 5 Most Interesting Cases…

Ever wonder what spies and scammers are doing with ChatGPT? It’s not just asking for five-paragraph essays, obviously.

OpenAI just dropped a wild new threat report (which we really want to call “Threat Level: Midnight”) detailing how threat actors from China, Russia, North Korea, and Iran are using its models for everything from cyberattacks to elaborate schemes, and it reads like a new season of Mr. Robot.

The big takeaway: AI is making bad actors more efficient, but it's also making them sloppier. By using ChatGPT, they’re leaving a massive evidence trail that gives OpenAI an unprecedented look inside their playbooks.

Here are the top 5 most fascinating examples of what they’ve caught:

1. North Korean-linked actors faked remote job applications. They automated the creation of credible-looking résumés for IT jobs and even used ChatGPT to research how to bypass security in live video interviews using tools like peer-to-peer VPNs and live-feed injectors.

2. A Chinese operation ran influence campaigns and wrote its own performance reviews. Dubbed “Sneer Review,” this group generated fake comments on TikTok and X to create the illusion of organic debate. The wildest part? They also used ChatGPT to draft their own internal performance reviews, detailing timelines and account maintenance tasks for the operation.

3. A Russian-speaking hacker built malware with a chatbot. In an operation called “ScopeCreep,” an actor used ChatGPT as a coding assistant to iteratively build and debug Windows malware, which was then hidden inside a popular gaming tool.

4. Another Chinese group fueled U.S. political division. “Uncle Spam” generated polarizing content supporting both sides of divisive topics like tariffs. They also used AI image generators to create logos for fake personas, like a “Veterans for Justice” group critical of the current US administration.

5. A Filipino PR firm spammed social media for politicians. “Operation High Five” used AI to generate thousands of pro-government comments on Facebook and TikTok, even creating the nickname “Princess Fiona” to mock a political opponent.

Why this matters: It’s a glimpse into the future of cyber threats and information warfare. AI lowers the barrier to entry, allowing less-skilled actors to create more sophisticated malware and propaganda. A lone wolf can now operate with the efficiency of a small team. This type of information will also likely be used to discredit or outright ban local open-source AI if we’re not careful to defend them (for their positive uses).

Now get this: The very tool these actors use to scale their operations is also their biggest vulnerability. This report shows that monitoring how models are used is one of the most powerful tools we have to fight back.

Every prompt, every code snippet they ask for help with, and every error they try to debug is a breadcrumb. They're essentially telling on themselves, giving researchers a real-time feed of their tactics. For now, the spies using AI are also being spied on by AI.

What’s a spy thriller without a great twist, right?

FROM OUR PARTNERS

StoryAI by Fullstory 

Meet StoryAI: your shortcut to smarter decision-making.

Powered by conversational AI agents, teams can analyze multiple user sessions, get instant answers to key questions, and predict behavior with confidence.

Turn insights into action and data into outcomes—faster than ever. The future of behavioral data starts right now—like, as soon as you click the button below. Just saying.

Check out the maturity matrix.

Prompt Tip of the Day

Here’s a genius prompt tip from Web Webster (of our TechnologyAdvice team):

Say you’re deep in an AI conversation, and you want to reference something it said three responses ago. So you type: “Hey, combine what you just wrote with that explanation about Plato and marketing from three responses ago.”

Unless you’re using Gemini 2.5 Pro, Claude Opus, or o3 / o4-mini-high, your model will undoubtedly lose its place and get confused across all your various drafts.

(related side note: Check out the new Apple paper below for more about why this happens).

Instead, try asking ChatGPT to use Cornell-style numbering (1.1, 1.2, 1.3, 2.1, etc) in its responses so you can reference specific sections later.

The prompt: 

Now you can say “combine section 1.2 with 2.1”, giving the AI’s attention mechanism more precise instructions for what you want. Worth adding “verbatim as written without changing anything” if you want exactly what 1.2 or 2.1 says, FYI!

Treats To Try.

*Anything marked with asterisks is sponsored content. Advertise in The Neuron here.

1. *Flow is the AI voice keyboard that turns speech into polished text in any iPhone app—Slack, iMessage, Gmail, Notion—5× faster than typing. Free plan with weekly word cap—try it here.

2. Mistral Code autocompletes your code and handles complex tasks like refactoring modules and updating tests (Mistral’s version of Codex / Claude Code).

3. Cursor’s latest update automatically reviews your GitHub pull requests with BugBot to catch bugs, runs Background Agents that code independently while you work, and remembers facts from your conversations to give better suggestions over time

4. Fieldy is an AI notetaker for real life conversations (wearable pendant).

5. Eleven v3 converts your text into expressive speech with controls for emotions, sound effects, and conversations between speakers in 70+ languages (demo).

6. Betterfeedback turns your surveys into natural conversations that ask smart follow-up questions.

7. You.com ARI scans 500+ sources and generates professional research reports that beat OpenAI Deep Research 76% of the time (raised $50M).

See our top 51 AI Tools for Business here!

Around the Horn.

• Gemini 2.5 Pro is now the top performing model in the world (according to benchmarks)—but Grok 3 and DeepSeek R1 are the highest intelligence for the price.

  • All along, Google was kinda always the frontrunner in the AI race. It’s probably worth listening to CEO Sundar Pichai and DeepMind CEO Demis Hassabis and their predictions about what comes next.

• Meta entered talks to invest $10B+ in the AI data-labeling startup Scale AI.

• Apple researchers released a new paper showing that “thinking” AI models like OpenAI's o1 and o3 hit a wall when problems get too complex—their accuracy drops to zero and they actually start thinking less, even though they have plenty of computing power available to keep trying.

  • Gary Marcus broke down why the Apple paper matters, calling it devastating evidence that these models can't reliably solve problems that seven-year-olds can, though he noted the paper has limitations since humans also struggle with complex versions of these puzzles.

  • For the counter point, Simon Willison argued that the last six months showed remarkable AI progress, with dramatic cost reductions, local models achieving near-flagship performance, and tools-plus-reasoning creating powerful new capabilities for practical applications.

• Apple has also had problems integrating language models over Siri—problems some insiders say they might not have faced if they started from scratch.

• Check out AI researcher + podcaster Lex Fridman’s experience testing out Google Beam’s new lifelike conferencing tool.

• It looks like Google’s AI Studio is switching to be “API based”, meaning it will no longer be free to use, likely by the end of the month (if not sooner).

• ICYMI: Apple and Anthropic are working on a “vibe-coding” tool… with WWDC coming up this week, we wonder if this will get a mention.

Monday Meme

A Cat's Commentary.