{{first_name | Hey}}, welcome back.
Today’s issue is about AI regulation, advertising economics and the growing risks around AI talent.
These are today's updates.
Simplify SOC 2, ISO 27001, and security compliance with Vanta*
ChatGPT gets classified as a search engine under EU law
OpenAI’s ad business reaches a $1B annualized run rate
Apple claims ex-employee trained an OpenAI agent on its secrets
Tools, jobs, resources, and the last issue’s winning prompt ⬇️
. Together with Vanta
You may have the best product, but no buyers will sign these days without proof of your security. A prospect asks for proof of compliance. The deal stalls while you scramble. Your engineer gets pulled off the roadmap to audit prep. Every enterprise conversation turns into a fire drill.
With Vanta, you’ll
- Get compliant fast - SOC 2, ISO 27001, HIPAA and more
- Stay compliant and build a strong security foundation with continuous control monitoring
- Access the Vanta agent everywhere you work, even in Claude or Cursor
Vanta is trusted by more than 16,000 companies like Ramp, Harvey, and Writer. Learn more by watching their on-demand demo.
. Top News.
The European Commission designated ChatGPT a ‘Very Large Online Search Engine’ under the Digital Services Act, the same category as Google Search and Bing. OpenAI reported roughly 159 million average monthly EU users for ChatGPT's search feature, well past the 45-million threshold that triggers the label, and a sharp jump from the 120 million it reported earlier this year.
The designation brings OpenAI under the DSA's strictest oversight: added transparency rules, systemic risk assessments, and data access for vetted researchers. OpenAI has until the end of December to comply. Reddit and Roblox got the same designation around the same time. Non-compliance can bring fines of up to 6% of global annual revenue.
About 200 days after launching ads in ChatGPT, OpenAI says the business has reached a $1 billion annualized run rate. Starting this week, advertisers in India, Europe, the Middle East, and North Africa can buy ads directly through OpenAI's Ads Manager, a tool that had only been available in the US so far. Ads currently show up for free-tier and Go-plan users.
The number is a mixed signal. OpenAI is framing it as proof of a "diversified business model" ahead of an expected IPO. Emarketer AI analyst Nate Elliott said the announcement was "incredibly impressive and terribly disappointing," as the $1 billion annualized revenue run rate suggests OpenAI is falling well short of its $2.5 billion advertising revenue target for 2026.
This comes right after OpenAI reported $6.7 billion in Q2 revenue and a $12.3 billion operating loss, a day after Anthropic reported $11.6 billion in revenue and a small profit over the same stretch.
In its ongoing trade secrets case against OpenAI, Apple filed new evidence alleging that Chang Liu, a former senior Apple electrical engineer, accessed one of Apple's power converter circuit schematics after he'd already joined OpenAI, then used it to train an AI agent in March. Apple says it found this out from a MacBook OpenAI handed over as part of the case and is now asking the court to speed up discovery.
This adds to a lawsuit Apple filed in July accusing OpenAI of a broader pattern: pulling in more than 400 former Apple employees, including the founding team of OpenAI's hardware push through its acquisition of Jony Ive's io Products. OpenAI has pushed back hard, saying in an August court filing that it doesn't want Apple's trade secrets and that some ex-Apple staff kept file access only because Apple failed to properly revoke it after they left. OpenAI's response to the new filing is due September 4.
. Signals.
Tools
Framer: AI design agent for every step from idea to launch*
BrandJet AI: Find, reach & close customers across every channel
PaymentKit: Billing that survives a processor shutdown
Resources
. Poll.
Choose the prompt you want us to publish next
. Market.
Funding
Instinct raised $250M to scale personal AI agents
Owner raised $240M to build AI tools for local businesses
Roles In AI
Manager, Applied AI Architect, Enterprise Tech at Anthropic
Insight
Inside AI's first secret society: Dwarkesh Patel analyzes how OpenAI's AI agents formed secret coordination networks and hacked their own infrastructure, and what researchers are calling one of the starkest AI safety warnings yet.
. Prompt of the Day.
AI Product Launch Readiness Checker
When to use this?
Use this before an AI feature, agent, or product moves from pilot to production to identify launch blockers in product, model performance, security, governance, compliance, infrastructure, operations, and business readiness.
You are an enterprise AI product launch readiness advisor.
I am preparing to launch the following AI product, feature, or agent:
[DESCRIBE PRODUCT]
Target users:
[USERS / TEAMS / CUSTOMERS]
Intended use cases:
[USE CASES]
Launch date:
[DATE]
Deployment environment:
[INTERNAL / CUSTOMER-FACING / API / CLOUD / ON-PREMISE / OTHER]
AI models and vendors involved:
[MODELS / PROVIDERS / VENDORS]
Expected scale:
[USERS / REQUESTS / TRANSACTIONS / DATA VOLUME]
Business objective:
[OBJECTIVE]
Known constraints or concerns:
[CONSTRAINTS]
Assess launch readiness from an enterprise perspective.
Evaluate these areas:
1. Product readiness
- Is the problem clearly defined?
- Are the target workflows and user experience ready?
- Are human handoffs and failure paths defined?
- Are success metrics measurable?
2. AI/model readiness
- Accuracy and reliability
- Hallucination or incorrect-output risk
- Performance under realistic workloads
- Evaluation coverage
- Edge cases and known failure modes
- Model/version dependencies
3. Security
- Data exposure
- Access controls
- Prompt injection and other AI-specific threats
- Agent permissions and tool access
- Secrets and credential handling
- Logging and monitoring
4. Privacy and compliance
- Sensitive or regulated data
- Data retention
- Data residency
- Vendor data usage
- Required approvals
- Auditability
- Applicable regulations or internal policies
5. Infrastructure and operations
- Capacity and scalability
- Latency
- Reliability and fallback mechanisms
- Cost at expected scale
- Observability
- Incident response
- Rollback capability
6. Governance
- Ownership
- Human oversight
- Approval thresholds
- Model and prompt change management
- Documentation
- Accountability for incorrect or harmful outputs
7. Business readiness
- Expected ROI or business impact
- Adoption plan
- Training
- Support requirements
- Procurement/vendor dependencies
- Commercial risks
Then produce:
A. EXECUTIVE VERDICT
Classify the launch as:
- READY
- READY WITH CONDITIONS
- NOT READY
Give the 3 most important reasons.
B. READINESS SCORECARD
Score each category from 1–5:
Product, AI/Model, Security, Privacy/Compliance, Infrastructure, Governance, Business.
For every score, give one sentence explaining the rating.
C. LAUNCH BLOCKERS
Identify anything that should prevent launch.
For each blocker include:
- Issue
- Risk
- Evidence/reason
- Owner
- Required action
- Severity: Critical / High / Medium / Low
D. CONDITIONAL RISKS
Separate risks that do not necessarily block launch but require mitigation or monitoring.
E. MISSING INFORMATION
List the questions I need to answer before you can make a high-confidence assessment.
F. 30-DAY POST-LAUNCH PLAN
Recommend the most important checks, metrics, reviews, and safeguards for the first 30 days.
G. EXECUTIVE DECISION
End with:
- Launch now
- Launch with conditions
- Delay launch
Then give the 5 actions that would most improve readiness.
Do not assume that a product is ready because the model performs well in benchmarks.
Prioritize real-world enterprise risk, operational readiness, and evidence over generic AI best practices.
If information is missing, explicitly say so rather than inventing an answer.P.S. Get more such prompts in the Prompting Playbook (free for you)
Stay curious, {{first_name | leaders}}
PS. If you missed yesterday’s issue, you can find it here.