🍭 OpenAI Wrote Its Own Safety Rules 📜

Good morning. OpenAI just published a document explaining how safe it is.

Definitely not because the EU's transparency deadline is in 63 days and there's an IPO to price.

"We were going to do this anyway."

Let’s dive in 👇

🍭 What’s Cookin’:

• OpenAI published its own safety rulebook

• CNN sued Perplexity over 17,000 scraped stories

• Groq raising $650M to reinvent itself

OpenAI
📜 Grades Its Own Homework

The Bite:

OpenAI published its Frontier Governance Framework, a public document mapping the company's internal safety and security practices onto two incoming regulatory frameworks: California's Transparency in Frontier AI Act and the EU AI Act's Code of Practice for General-Purpose AI.

The framework covers risk assessments across cyber offense, CBRN threats, harmful manipulation, and loss-of-control scenarios, using a tiered system to classify model capabilities from low to catastrophic.

EU compliance sits with OpenAI Ireland Limited. The full transparency rules take effect August 2, 2026. OpenAI's Q4 IPO timeline runs in parallel.

Snacks:

• The framework defines "systemic risk" as scenarios involving 50+ fatalities or $1B+ in property damage from a single incident

• A Tier 3 cyber offense model can autonomously find and exploit zero-days across hardened systems without human help

• A Tier 3 loss-of-control model can evade its own chain-of-thought monitoring and operate autonomously for sustained periods

• OpenAI Ireland Limited holds EU compliance responsibility; OpenAI OpCo LLC handles California's TFAIA

• The company commits to reassessing the framework every 12 months and updating safety reports every 6 months for its most capable models

Why it Bites:

The timing tells you everything.

OpenAI published this framework, not because they're chilling and reflective, but because the EU AI Act's full transparency rules land on August 2, and the company is trying to price an IPO this year.

That's two audiences with one document.

Regulators get a compliance artifact they can point to.
Investors get a signal that the regulatory risk is "managed."

The framework itself is thorough: tiered risk categories, defined thresholds, external auditing commitments, incident response protocols.
On paper, it's exactly what responsible frontier AI governance looks like.

But here's the thing: OpenAI wrote its own tiers, defined its own thresholds, and will assess its own models against them.

The external experts it cites are "solicited" by OpenAI, not appointed by a regulator. The 12-month review cycle is self-imposed and self-enforced.

So this is their play: publish a document detailed enough that it looks like regulation, before actual regulation arrives with teeth.

If the EU accepts OpenAI's framework as a credible baseline, every other frontier lab is now implicitly expected to match it. The company that needs compliance the most just got to write the template everyone else will be judged against.

The rulebook is real.

The question is who's holding the pen, and who's holding the grade sheet.

Plaid
📈 AI is raising the bar in finance

Over half of Americans have used AI to manage their finances.
And they're not going back. 

The State of Intelligent Finance, Plaid's latest report in partnership with the Harris Poll, surveyed 2,000+ American adults on how AI is changing what people expect from financial products.

Consumers want products that understand their situation, guide their decisions, and act on their behalf. Trust hinges on transparency and accountability.

The report covers:

• Where consumers are most ready to let AI take action?

• What builds trust and what breaks it?

• How teams can turn rising expectations into a competitive advantage?

_ToolBox™
🧰 5 BRAND NEW AI LAUNCHES

🎞️ Clipto
Search years of videos, meetings, and audio by describing a moment like “that client call where Sarah mentioned pricing” and actually find it.

🧠 Second Brain for AI
Gives ChatGPT, Claude, and Cursor a memory that survives past a single conversation, so you stop re-explaining your life every Tuesday.

🔒 TabTasker
Edit PDFs, transcribe audio, process images, and knock out 50+ tasks entirely offline, because not every file belongs on someone else's server.

📚 Web Clipper for NotebookLM
Turn NotebookLM into a research beast by saving anything from Reddit threads to YouTube playlists and exporting insights wherever you work.

🗣️ Copycat Cafe
Learn a language by mimicking real native speakers, then get roasted (politely) by AI until your pronunciation stops sounding like Google Translate with a head cold.

_{Everything Else}
🧠 You Need to Know

⚖️ CNN Sues Perplexity Over 17,000 Scraped Stories
→ The lawsuit alleges Perplexity scraped thousands of CNN stories, photos, and videos to power its products. It's the first TV network to sue an AI company.

🔍 DuckDuckGo Installs Surge After Google's AI Search Overhaul
→ U.S. app installs jumped 18% week-over-week after Google I/O, peaking at 30% on Memorial Day.

⚙️ Groq Raises $650M as It Pivots Away From Being a Chipmaker
→ Existing backers Disruptive and Infinitum are backstopping the round as Groq reinvents itself under new leadership after its $20B Nvidia licensing deal took the founder and most senior engineers.

📜 OpenAI Publishes Frontier Governance Framework
→ The document maps OpenAI's internal safety practices onto California's Transparency in Frontier AI Act and the EU AI Act's Code of Practice.

📈 Salesforce Agentforce Hits $1.2B on Annual Recurring Revenue
→ Q1 FY2027 revenue came in at $11.13B with Agentforce crossing the $1B ARR mark for the first time.