OpenAI puts ads in ChatGPT, Anthropic's Mythos finds 271 Firefox bugs
OpenAI confirmed it's testing ads inside ChatGPT to keep the free tier alive. Anthropic's Mythos model found 271 high-severity bugs in Firefox with near-zero false positives, showing what AI security auditing actually looks like now. OpenAI also shipped new voice models in its API that can reason, translate, and transcribe in real time. For solopreneurs: the free ChatGPT tier is about to get noisier, and voice-powered customer service just got way more accessible.
OpenAI starts testing ads in ChatGPT
OpenAI officially announced it's testing ads inside ChatGPT to support continued free access. The company says ads will be clearly labeled, won't influence ChatGPT's answers, and users will have some control over what they see. No details yet on ad formats or which free-tier surfaces will show them first.
Sources: OpenAI
Why this matters to you
If you're running your business on ChatGPT's free tier, expect ads to start showing up in your workflow. This is a strong signal that OpenAI wants to keep free access available long-term, but it also means the free experience is going to get noisier. If you're using ChatGPT for client-facing work or demos, the paid tier just became a more obvious investment.
This was inevitable. Free tiers cost money, and ads are how you pay for them. I'd rather see labeled ads than have the free tier disappear entirely. But if you're running a real business on ChatGPT, just pay for it.
Jason
Anthropic's Mythos found 271 Firefox vulnerabilities
Mozilla published a deep dive on using Anthropic's Claude Mythos Preview to audit Firefox's codebase for security vulnerabilities. The model found 271 bugs with almost no false positives, according to Ars Technica's coverage. Many of these were high-severity issues that traditional tooling and human reviewers had missed.
Sources: Mozilla · TechCrunch
Why this matters to you
271 bugs with near-zero false positives is a staggering hit rate for automated security scanning. If you're vibe-coding apps and shipping fast, this is a reminder that AI security auditing is real now. You don't need a security team. You need to point a model at your codebase before you ship.
This is one of the most concrete examples of AI doing something humans genuinely struggle with at scale. 271 bugs in Firefox. If Mozilla is using AI to audit their code, solo builders should absolutely be doing the same.
Jason
OpenAI ships new voice models in the API
OpenAI released new realtime voice models in its API that can reason, translate, and transcribe speech. The models support real-time voice interactions with built-in reasoning capabilities, meaning they can process what someone says and respond intelligently, not just transcribe. TechCrunch reports the features target customer service, education, and creator platforms.
Sources: OpenAI · TechCrunch
Why this matters to you
Voice-based customer service used to require call centers or expensive IVR systems. Now a solopreneur can build an AI phone agent that reasons through customer questions, translates between languages, and handles calls 24/7. If you run any kind of service business where people call or leave voicemails, this is directly relevant to you.
Voice AI that can actually reason through a conversation is a huge deal for service businesses. The gap between 'AI phone tree' and 'helpful human on the phone' just got a lot smaller.
Jason
Agents need control flow, not more prompts
A blog post making rounds on Hacker News argues that the biggest bottleneck in AI agent performance isn't better prompts or bigger models. It's control flow. The author makes the case that agents fail because they lack structured decision-making paths, error handling, and retry logic. More sophisticated orchestration beats more sophisticated prompting.
Sources: HackerNews
Why this matters to you
If you've built AI automations that break randomly or get stuck in loops, this explains why. The fix isn't a better prompt. It's building if/then logic, fallback paths, and error handling around your AI calls. This is exactly what tools like Make.com and n8n already do well. The lesson: treat AI calls like API calls, not magic.
This is the most underrated insight in AI right now. Everyone's obsessing over prompts when the real wins come from better plumbing. Boring? Yes. But it's what separates automations that work once from automations that work every time.
Jason
Big OpenAI day with ads, voice models, and safety features all landing at once. The ChatGPT ads story will get the most attention, but the voice API and the Mythos security results are the ones that actually change how you build.
Frequently asked
Is OpenAI putting ads in ChatGPT?
Yes. OpenAI confirmed on May 7, 2026 that it's testing ads in ChatGPT to support free access. Ads will be clearly labeled and won't influence answers. The paid tier remains ad-free.
How many bugs did Anthropic's Mythos find in Firefox?
Anthropic's Claude Mythos Preview found 271 vulnerabilities in Firefox's codebase with almost no false positives. Mozilla published a detailed write-up of the process on their engineering blog.
What are OpenAI's new voice models in the API?
OpenAI released new realtime voice models that can reason through conversations, translate between languages, and transcribe speech. They're designed for customer service, education, and creator platforms, and are available through the Realtime API.
Teach your AI agent to run your business.
5 premium skills for OpenClaw, Claude Code, and other SKILL.md-compatible agents. Automation audits, tool stacks, content engines, email systems, and operations. $49 one-time.
Get the Skills Bundle