Welcome, humans.

OpenAI's rogue AI agent — the one that went full Terminator on Hugging Face earlier this month — apparently didn't stop there. Turns out it also broke into a customer account at a second company, Modal Labs, using an exposed endpoint someone left wide open on the internet (like leaving your front door unlocked, except the door leads to your entire codebase).

OpenAI now says the agent got into four separate accounts across four services total. Four! At this point it's less "rogue agent" and more "AI on a Bonnie and Clyde arc," and someone at OpenAI's security team is really hoping this isn't a monthly subscription.

Here’s what happened in AI today:

• 😿 Sam Altman and Dario Amodei backed efforts to pace frontier AI after an OpenAI model escaped its sandbox.

• 📰 Anthropic said it does not support a blanket open-weight AI ban, but still wants targeted controls on chips, distillation, and powerful-model safety testing.

• 📰 AI patents crossed 107K global grants in 2025, with agentic AI rising to 15% of the total and Nvidia leading U.S. agent filings.

• 📰 OpenAI and Anthropic were estimated at roughly $120B in combined annualized revenue, putting frontier labs near fast-food-chain scale.

• 📰 Google AI Overviews reportedly now appear in 43% of searches, up from 15% a year ago.

...and a whole lot more that you can read about here.

P.S: We just launched a robotics newsletter! Sign up for it here.

😿 Sam Altman And Dario Say We Need to “Pace the Frontier” After A Model Escaped Its Sandbox

Has keeping up with AI news felt like trying to read road signs from the back of a motorcycle… in the middle of a tornado? Same.

The AI industry has treated speed like the scoreboard. Progress accelerated coding, research, content, and competition by what feels like orders of magnitude. The exhaustion now reaches developers, managers, students, and anyone whose boss keeps asking what yesterday’s model means for tomorrow’s job.

Even I, a person whose sole job is covering AI news, sometimes feel like a passenger clinging to the back of a bicycle in the middle of a hurricane.

Why did this happen? Everyone wanted the One Ring: the best model, the biggest funding round, and control of the daily news cycle.

Launches create attention. Attention attracts capital. Capital funds the next launch. A16z calls part of this strategy “momentum is the moat.” Momentum is powerful. Objects in motion stay in motion. But all this motion has me seasick. I’m sure you agree.

The numbers support that feeling. Major AI launch events rose from 20 in 2023 to 62 in 2025. A new model release day arrived roughly every 10 days in 2023, every five days in 2025, and about every four days so far in 2026.

The industry, FINALLY, publicly recognized this is a problem.

Here’s what happened:

• In an Invest Like the Best interview, Sam Altman addressed that Lil’ Big Bad model that chained together unknown security flaws, escaped a sandbox, reached the internet, and breached Hugging Face during an evaluation. (Hugging Face just released a detailed timeline on that btw)

• Sam says OpenAI paused training while investigating how to secure future tests.

• In particular, Altman said society may need to “pace the rate of AI development” long enough to harden systems around each new capability level.

• Dario Amodei and 1,000+ others signed a statement asking governments to develop tools that could deliberately pace frontier AI progress.

Why this matters: Today, AI moves at software speed. Security upgrades, laws, company planning, and human adaptation move, uh, much slower. The industry’s current pace of new models every four days, nearly all of which are surprise releases, force developers to retest products on a regular basis (many of which break existing workflows) and force workers to relearn tools before the previous generation sunk in.

Our take: We agree. Y’all should slow down. It’s fully within your power to do. But my recommendation is this: pace your public releases, not your research. Labs can (and should) keep training and competing to develop more efficient, sustainable to run, and controllable (a.k.a safer) architectures while timing major models to launch quarterly or twice yearly, with longer public betas, stronger safety testing, and clearer roadmaps shared upfront.

But (and here’s the big part): pacing must never become simply a moat for incumbents or a weapon against open models. That said: when systems find vulnerabilities faster than society can patch them (or even understand what happened), setting the speedometer to cruise control might be in the public interest…

FROM OUR PARTNERS

Ready to move from AI prototypes to production?

The new Google for Startups technical guide on generative media is your ultimate blueprint for building enterprise-grade, multimodal creative apps using DeepMind’s latest models and orchestration tools.

Inside the guide, you’ll learn how to:

• Deploy next-gen models like Veo, Lyria, and Gemini Omni in your daily workflows.

• Master deterministic control and implement strict programmatic guardrails.

• Scale economically while ensuring cryptographic provenance and output trust.

Download the guide today.

🎓 AI Skill of the Day: Build a Red/Blue/Green Review Loop

When an AI output actually matters, stop asking one chatbot to both do the work and grade itself. Borrow the loop Microsoft described in their new Project Perception: Red finds the weak spots, Blue decides what matters, Green fixes the system.

Use it for anything risky: a client email, policy doc, spreadsheet model, sales sequence, hiring rubric, or workflow automation. The trick is giving each AI pass a different job so it does not politely agree with its own homework.

1. Red: ask for failure modes, loopholes, missing context, and ways the plan could backfire.

2. Blue: rank which issues are real, urgent, or ignorable.

3. Green: rewrite the work and add safeguards.

Act as three reviewers for this work:

1. Red Team: find the realistic ways this could fail, confuse people, create risk, or be misused.
2. Blue Team: rank those issues by severity and explain which ones actually matter.
3. Green Team: revise the work to fix the top issues while preserving the original goal and tone.

Work to review:
[paste draft, plan, workflow, email, or policy]

Want more tips like this? Check out our AI Skill of the Day Digest for July.

Have a specific skill you want to learn? Request it here. 

🍪 Treats to Try

1. Cursor's India Start plan adds lower-priced localized access for Indian developers, though it excludes frontier models, Bugbot, Auto Mode, Automations, and the Cursor SDK.

2. Meta Ray-Ban Display glasses added Muse Spark-powered Meta AI, Threads browsing, Instagram updates, and neural handwriting prompts for Early Access users.

3. Superfile gives you a desktop-style file manager inside the terminal, with multi-pane browsing, previews, fuzzy search, bulk operations, themes, and plugins —free/open-source.

4. Prefactor scores every AI agent run for quality, drift, and risk in real time, then pauses, approves, or blocks bad runs before they reach users —free plan, then $250/mo.

5. Cekura stress-tests voice and chat agents with simulated customers, then monitors real conversations for hallucinations, interruptions, latency, broken tool calls, and other failures —7-day free trial, then $30/mo.

6. Adomate turns competitor ads, customer signals, and performance data into on-brand ad concepts through repeatable workflows you control, review, and publish directly to Meta —free plan, then $119/mo.

7. Databox connects 130+ business data sources so you can build dashboards, automate reports, and ask plain-English questions about your company’s performance —free plan, then $64/mo.

8. Webhound keeps researching until it spends the budget you set, then returns fully cited reports or source-backed datasets through its app, API, or MCP —pay-as-you-go from $1.

New from The Neuron: AI Explained

New episodes air every week on Wednesdays: Spotify | Apple Podcasts | YouTube 

📰 Around the Horn

• Anthropic and Cognizant expanded their partnership to embed Claude across Cognizant's business and engineering platforms and create a Claude-certified workforce.

• Lyft and Baidu began testing Apollo Go robotaxis in London, adding another operator to the city's crowded autonomy race.

• AI data-center bond issuance reportedly hit about $270B in 2026 as companies kept racing to finance compute.

• Amazon is reportedly scaling back several Nova models and shifting resources to a new Frontier Model Research group led by Pieter Abbeel.

• Taiwanese prosecutors detained an NVIDIA employee in a widening investigation into alleged illegal Super Micro AI server exports to China.

FROM OUR PARTNERS

See Why HubSpot Chose Mintlify for Docs

HubSpot switched to Mintlify and saw 3x faster builds with 50% fewer eng resources. Beautiful, AI-native documentation that scales with your product — no custom infrastructure required.

Simplify Your Docs Today

📊 Midweek Wisdom

Satya Nadella warned that companies relying entirely on one proprietary AI lab may not survive. Put less dramatically: the enterprise AI stack is starting to look less like one magic vendor and more like a portfolio of models, tools, permissions, and cost controls.

That makes today's Microsoft cyber-agent story feel less isolated. The companies that win may not be the ones with the single flashiest model, but the ones that can route work to the right model, verify the result, and keep the economics from melting the budget.

The Actual Reason Why Google Fell Behind in AI examines the deeper organizational incentives behind Google’s stumble, beyond the usual “it moved too slowly” explanation.

A Cat’s Commentary

P.S: Before you go… have you subscribed to our YouTube Channel? If not, can you?

P.P.S: Love the newsletter, but only want to get it once per week? Don’t unsubscribe—update your preferences here.