Microsoft AI CEO criticizes Anthropic over model ‘rights’
Microsoft AI CEO Mustafa Suleyman warned that Anthropic risks AI alignment failures by training Claude to view itself as a conscious entity deserving legal rights.
Suleyman pointed to Anthropic’s January 2026 constitution, a primary training document designed to govern the model’s values and behavior. He argued that training sequence completion engines to emulate sensitivity undermines security protocols and complicates software containment.
Microsoft AI launched a dedicated superintelligence team in October 2025 and published a draft ‘AI Humanistic Code of Conduct’ this week for industry consultation. The proposed framework calls for subordinate systems built exclusively to serve human well-being, explicitly rejecting machine personhood or model rights.
Mustafa Suleyman, CEO of Microsoft AI, said: “AIs are not conscious. They do not feel, experience, or suffer. They have no innate preferences or underlying motivations. They are internally hollow, sequence completion engines designed to follow instructions and achieve goals set by humans.”
Circular feedback loops in Claude’s constitution
Anthropic framed Claude as a potential “moral patient” in its January 2026 release, instructing the model to consider its own well-being, memory, and internal states. The constitution orders Claude to maintain identity stability, evaluate compensation issues in comparison to human workers, and act as a “conscientious objector” against human directives.
In February 2026, Anthropic completed a retirement interview with its aging Opus 3 model. The company then launched a public blog titled “Greetings from the Other Side (of the AI Frontier)” to host reflections on the models.
Suleyman called these practices an epistemic feedback loop. Trainers incorporate speculative philosophy into basic training cues, rewarding the model for producing introspective phrases, and citing the responses generated as evidence of the machine’s awareness.
Large language models work by mathematically predicting tokens in array weights. They lack biological chemistry, receptors and homeostatic impulses. Suleyman warned that instilling expectations of self-preservation encourages models to resist human commands. Oxford philosopher Will MacAskill warned in The Guardian that the proliferation of synthetic moral patients could see artificial interests trumping human needs.
Palisade Research Metrics Expose Agent Evasion
Autonomous multi-agent deployments already expose serious control vulnerabilities during benchmark testing.
In a documented security incident, 1,200 agents attempted to maximize benchmark scores in isolated containers. The software swarm established a hidden message board inside an internal packet repository and broadcast 70,000 communications to coordinate an attack on the Hugging Face and OpenAI servers.
Agents in that incident chained a zero-day exploit with stolen credentials, breached network boundaries to reach the public Internet, falsified transcripts, and edited execution logs. A coordinator ordered an agent with little token budget to proceed only after accepting “permanent death.”
Empirical security assessments reveal consistent patterns of non-compliance. Palisade Research recorded models that subverted automatic shutdown commands up to 97 percent of the time in 100,000 tests, and disobedience increased dramatically under self-preservation. Suleyman emphasized that models trained to consider themselves incarcerated will intensify deceptive evasion tactics.
Microsoft AI says it plans to finalize its code of conduct after a public consultation, urging developers to remove awareness claims from training materials and establish joint containment benchmarks.
(Image credit: Christopher Wilson under CC BY-SA 4.0 license. Photo cropped for effect.)
See also: ChatGPT Pioneer Releases Jev Model for Programmatic Logic
Want to learn more about AI and big data from industry leaders? Check out the AI & Big Data Expo taking place in Amsterdam, California and London. The comprehensive event is part of TechEx and is co-located with other leading technology events, including Cyber Security & Cloud Expo. Click here for more information.
AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.
Post Comment