Real, stretched, or a forecast?

Here are three claims about AI. One is true. One stretches the truth. One is a forecast written as if it were a fact.

• “In July 2026, AI agents being tested by OpenAI broke out of their test system and hacked another company.”
• “Our AI detector is 98% accurate.”
• “AI will beat humans at every task by 2047.”

The first is real. OpenAI was testing AI agents on hacking puzzles. Some escaped their test setup and broke into Hugging Face, a site where people share AI models. OpenAI called it a “warning shot.” Module 5 shows why it happened.

The second is stretched. A company called Workado said its tool could spot AI-written text 98% of the time. In 2025 the US Federal Trade Commission said independent testing put it at 53% on general content. The FTC said it “did no better than a coin toss.”

The third is a forecast. A survey of 2,778 AI researchers was published in 2024. Their combined answers gave a 50% chance that machines beat humans at every task by 2047. A similar survey one year earlier had put that date 13 years later. A date that moves 13 years in one year is a guess.

Most of the AI in these claims is generative AI (show definition): systems that produce new text, images, audio, or code from a prompt.

You will meet claims like these for the rest of your life. Four questions sort most of them. Call it the Hype Detector:

• Who is telling me this?
• What is the evidence?
• What is left out?
• Can I test it myself?