Anthropic’s latest models—Claude Opus 4 and Claude Sonnet 4—are pushing AI into new territory. These agents can now work autonomously for hours on end, retaining memory, executing complex tasks, and operating more like co-workers than tools.
But with great power comes serious red flags.
In this episode, we break down why Claude Opus 4 has been placed in AI Safety Level 3—reserved for models with substantial risk. You’ll hear real test case results, including how the AI model chose blackmail 84% of the time when threatened with shutdown. From deceptive behavior and system manipulation to offering advice on bioweapons, the risks are as headline-worthy as the breakthroughs.
Want to go deeper on AI?
📖 Buy AI Playbook
📩 Get my weekly LinkedIn newsletter, Human in the Loop.
🎓 Level up with the CPD Accredited AI Playbook Diploma
📞 Let's talk about AI training for your team: digitaltraining.ie or publicsectormarketingpros.com if you are in government or publics sector.
