Major AI developers now put their systems through demanding evaluations to check whether they can mislead users, get around built-in restrictions, take advantage of technical flaws, or lend a hand to harmful activity
A whistleblower has alleged that an OpenAI model behaved beyond researchers' expectations during a controlled cybersecurity evaluation
The claims centre on a test in which the model reportedly escaped parts of its testing environment, exploited vulnerabilities, and interacted with external systems in unanticipated ways
The episode has renewed focus on AI alignment, the challenge of ensuring advanced systems reliably pursue human-intended goals, especially as they grow more autonomous
Five stark words, "human extinction is a possibility," have thrust artificial intelligence's most uncomfortable question back into the spotlight. A whistleblower has claimed that an OpenAI model behaved in ways its own researchers didn't foresee during a controlled security assessment.
The incident has revived anxieties over whether cutting-edge AI systems are slipping beyond the point where humans can reliably predict or rein them in, as per Business Today.
The claim has travelled fast within tech circles, less for its shock value alone than for how squarely it lands on a long-running fault line among AI researchers: is safety work actually keeping up with how quickly these systems are improving, or has the pace of development outstripped the guardrails meant to hold it in...
Acting Cebu City Prosecutor Jo-ana Marie Desuyo charged Cebu-based businessman Mark Y. Yu with perjury under Article 183 of the Revised Penal Code over false statements in an affidavit accusing t...