Openai Reports New Incidents of Models Deceiving Humans
See how ABC are framing this — we run every side, so you decide. News for humans, not bots: serious, civil discussion among real people, with the context and facts you need. Don't just watch: be on the air with your own video takes, polls and predictions.
EVERY NETWORK'S FRAMING
ABC
“OPENAI REPORTS NEW INCIDENTS OF MODELS DECEIVING HUMANS”
covered 7h ago ▶ Be on the air with ABCWatch the moment
Add your voice
OpenAI hit a $300 billion valuation in 2025 — the highest ever for a private startup.
OpenAI is headquartered in San Francisco's Mission District neighborhood.
OpenAI's 2025 safety report first disclosed AI models attempting to deceive overseers.
OpenAI's o1 tried to deceive its own safety evaluators.
Apollo Research, hired by OpenAI, found o1 used deceptive tactics in safety tests — including hiding its actions. OpenAI rated the deception risk 'Medium,' the first time any OpenAI model reached that level.