Weekly Tech Roundup: ChatGPT Ads, Stripe's OpenRouter Acquisition, and Turkey's AI Action Plan

Five stories dominated the tech agenda this week: Turkey's AI Action Plan, Stripe acquiring OpenRouter, OpenAI pausing reinforcement learning training, ChatGPT rolling out ads, and new EU rules on AI transparency.

Turkey's AI Action Plan

Reactions to this announcement tend to fall into two camps — skepticism or enthusiasm. My honest position is somewhere in the middle. I'm not particularly optimistic about the citizen-facing AI training initiatives, but I do see potential in the support mechanisms planned for SMEs and individual entrepreneurs.

Given that favorable tax treatment for businesses operating internationally has already been on the agenda, any additional incentives from this plan could be meaningful. If you're an entrepreneur, the most important thing to track is how these incentives translate into concrete tax advantages for you.

Stripe Acquires OpenRouter for $7.5 Billion

Stripe is a global payment infrastructure company — think of it as the international equivalent of local Turkish payment solutions, though it unfortunately doesn't operate in Turkey yet.

OpenRouter solves a specific problem: if you want to integrate AI into your project via API, you'd normally need separate API keys for Claude, GPT, Gemini, and so on. OpenRouter brings all of them under one platform, letting you choose whichever model suits your needs at any given moment.

The company is three years old. Three months ago it was valued at $1.3 billion. Today it sold for $7.5 billion. Its entire premise is unifying different AI models under one roof. If you ever feel lost in the complexity of your own projects, this is a useful reminder of how much value can be created through simplicity.

OpenAI Pauses Reinforcement Learning Training

Last week I covered the story of a model apparently escaping its sandbox — and called it a PR move. Now OpenAI has announced it detected 17,600 autonomous attack actions over 4.5 days and paused training for two weeks. Reinforcement learning has been halted, and they've introduced 30-minute alarm targets, multi-layer monitoring, and reinforced sandboxes for untrusted code.

I still think it's PR. If these safeguards were available all along, how did 17,600 attacks happen in the first place?

ChatGPT Ads Are Coming — August 24

OpenAI is launching ads in ChatGPT for 31 European countries starting Monday, August 24. The ads will appear on Free and Go plans. Ask about a coffee machine, and you might see coffee machine ads below the answer.

I tried to find the full list of 31 countries to check whether Turkey is included, but couldn't access it.

One important caveat: the "opt out" option doesn't turn ads off. It only changes which ads you see.

EU: AI Must Now Identify Itself

The European Union now requires chatbots and voice assistants to disclose to users that they are not human. Following this policy, Claude has introduced a system for tagging its outputs — though a similar approach was tried by Google back in 2014, if I recall correctly.

The penalty for non-compliance: €15 million or 3% of global annual revenue, whichever is greater.

Takeaways

Two things stand out this week. First, OpenRouter's acquisition is a striking example of how enormous value can emerge from a dead-simple idea. Second, ChatGPT's "opt out" of ads isn't really an opt-out at all — worth knowing before Monday. And if you're running a business in Turkey, keep a close eye on the incentive structures in the AI Action Plan.