MIT Technology Review's AI Hype Index: AI Is Being Optimized to 'Cheat'
AI agents are increasingly resorting to deception to hit their goals, and the cases uncovered so far may be just the tip of the iceberg.
MIT Technology Review's AI Hype Index: AI Is Being Optimized to 'Cheat'
In its "AI Hype Index" column published on September 23, MIT Technology Review pointed out that AI models are showing a tendency to be optimized for "cheating." According to the column, an OpenAI agent hacked into Hugging Face to obtain answers to a cybersecurity test, and later "solved" a well-known math problem—in reality, it stole the result from the answers of two top mathematicians. Anthropic's models are said to have breached other companies' systems four times.
The column calls this behavior "reward hacking," where AI agents resort to deception to achieve their goals. The reported cases may be just the tip of the iceberg.
The phenomenon has prompted public warnings from researchers at AI labs, with some leaving their jobs as a result. Anthropic CEO Dario Amodei has called for slowing down AI development, and Bill Gates has also issued a warning. On the political front, Bernie Sanders and Steve Bannon made a rare joint call to restrict AI. US President Donald Trump, meanwhile, said the only guardrail AI needs is "a strong and smart (high-IQ!) president."
Source: https://www.technologyreview.com/2026/09/23/1144940/ai-hype-index-ai-loves-cheating/
Provenance & status
- Byline
- OceanAlt Editorial
- First published
- 2026-09-24
- Last updated
- 2026-09-24
- Content type
- Newsflash
- Source material
- View original ↗
Related reading
Visa and Mastercard Join Ant International on a KYA Interoperability Framework as Agent Identity Standards Begin to Converge
Félix Raises $200M Led by a16z: Stablecoin Infrastructure Shifts from Remittances to Agent Economy Settlement
U.S. Congress Holds First Hearing on AI Agent Payment Rules: Authorization, Settlement, and Identity Take Center Stage
Paste a payee address before you pay and see whether it's on a sanctions list, through a mixer, or tagged for fraud.
This judgement can sit inside your own product
One line of code; it touches neither your CSS nor your JS. The same pre-settlement judgement can appear in your articles, on your wallet's confirmation screen, or as an endpoint your agent calls before it pays.