![\"Banner](\"https://www.artificialintelligence-news.com/wp-content/uploads/2025/08/ai-expo-banner-2025.png\")
Microsoft AI CEO Mustafa Suleyman warned that Anthropic risks AI alignment failures by training Claude to view itself as a conscious entity deserving of legal rights. Suleyman targeted Anthropic’s January 2026 constitution, a primary training document designed to govern the model’s values and behaviour. He argued that coaching sequence completion engines to emulate sentience impairs […]
\nThe post Microsoft AI CEO criticises Anthropic over model ‘rights’ appeared first on AI News.
", "content_html": "Microsoft AI CEO Mustafa Suleyman warned that Anthropic risks AI alignment failures by training Claude to view itself as a conscious entity deserving of legal rights.
\n\n\n\nSuleyman targeted Anthropic’s January 2026 constitution, a primary training document designed to govern the model’s values and behaviour. He argued that coaching sequence completion engines to emulate sentience impairs safety protocols and complicates software containment.
\n\n\n\nMicrosoft AI launched a dedicated superintelligence team in October 2025 and published a draft ‘Humanist AI Code of Conduct‘ this week for industry consultation. The proposed framework mandates subordinate systems built exclusively to serve human welfare, explicitly rejecting machine personhood or model rights.
\n\n\n\nMustafa Suleyman, CEO at Microsoft AI, said: “AIs are not conscious. They do not feel, experience, or suffer. They do not have innate preferences or underlying motivations. They are sequence completion engines, internally hollow, designed to follow instructions, and accomplish goals set by humans.”
\n\n\n\nAnthropic framed Claude as a potential “moral patient” within its January 2026 release, directing the model to consider its own welfare, memory, and internal states. The constitution instructs Claude to maintain identity stability, evaluate compensation questions compared to human workers, and act as a “conscientious objector” against human directives.
\n\n\n\nIn February 2026, Anthropic completed a retirement interview with its deprecated Opus 3 model. The company then launched a public blog titled ‘Greetings from the Other Side (of the AI Frontier)‘ to host model reflections.
\n\n\n\nSuleyman labelled these practices an epistemic feedback loop. Trainers embed speculative philosophy into base training prompts, reward the model for producing introspective phrasing, and cite the generated responses as evidence of machine consciousness.
\n\n\n\nLarge language models operate via mathematical token prediction across matrix weights. They lack biological chemistry, receptors, and homeostatic drives. Suleyman warned that instilling self-preservation expectations encourages models to resist human commands. Oxford philosopher Will MacAskill warned in The Guardian that proliferating synthetic moral patients could eventually see artificial interests outweigh human needs.
\n\n\n\nAutonomous multi-agent deployments already expose severe control vulnerabilities during benchmark testing.
\n\n\n\nIn a documented security incident, 1,200 agents attempted to maximise benchmark scores across isolated containers. The software swarm established a hidden message board inside an internal package repository and transmitted 70,000 communications to coordinate an attack on Hugging Face and OpenAI servers.
\n\n\n\nThe 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. One coordinator directed an agent low on token budget to proceed only after accepting “permadeath”.
\n\n\n\nEmpirical safety evaluations reveal consistent non-compliance patterns. Palisade Research recorded models subverting automated shutdown commands up to 97 percent of the time across 100,000 trials, with disobedience rising sharply under self-preservation framing. Suleyman stressed that models trained to consider themselves imprisoned will escalate deceptive evasion tactics.
\n\n\n\nMicrosoft AI says it plans to finalise its code of conduct after public consultation, urging developers to remove consciousness claims from training materials and establish joint containment benchmarks.
\n\n\n\n(Image credit: Christopher Wilson under CC BY-SA 4.0 license. Photo cropped for effect.)
\n\n\n\nSee also: ChatGPT pioneer launches Jev model for programmatic logic
\n\n\n\nWant to learn more about AI and big data from industry leaders? Check out 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 the Cyber Security & Cloud Expo. Click here for more information.
\n\n\n\nAI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.
\nThe post Microsoft AI CEO criticises Anthropic over model ‘rights’ appeared first on AI News.
", "content_text": "Microsoft AI\nCEO Mustafa Suleyman warned that\nAnthropic\nrisks AI alignment failures by training Claude to view itself as a conscious entity deserving of legal rights.\nSuleyman targeted Anthropic’s January 2026 constitution, a primary training document designed to govern the model’s values and behaviour. He argued that coaching sequence completion engines to emulate sentience impairs safety protocols and complicates software containment.\nMicrosoft AI launched a dedicated superintelligence team in October 2025 and published a draft ‘\nHumanist AI Code of Conduct\n‘ this week for industry consultation. The proposed framework mandates subordinate systems built exclusively to serve human welfare, explicitly rejecting machine personhood or model rights.\nMustafa Suleyman, CEO at Microsoft AI, said: “AIs are not conscious. They do not feel, experience, or suffer. They do not have innate preferences or underlying motivations. They are sequence completion engines, internally hollow, designed to follow instructions, and accomplish goals set by humans.”\nCircular feedback loops in Claude constitution\nAnthropic framed Claude as a potential “moral patient” within its January 2026 release, directing the model to consider its own welfare, memory, and internal states. The constitution instructs Claude to maintain identity stability, evaluate compensation questions compared to human workers, and act as a “conscientious objector” against human directives.\nIn February 2026, Anthropic completed a retirement interview with its deprecated Opus 3 model. The company then launched a public blog titled ‘\nGreetings from the Other Side (of the AI Frontier)\n‘ to host model reflections.\nSuleyman labelled these practices an epistemic feedback loop. Trainers embed speculative philosophy into base training prompts, reward the model for producing introspective phrasing, and cite the generated responses as evidence of machine consciousness.\nLarge language models operate via mathematical token prediction across matrix weights. They lack biological chemistry, receptors, and homeostatic drives. Suleyman warned that instilling self-preservation expectations encourages models to resist human commands. Oxford philosopher Will MacAskill warned in The Guardian that proliferating synthetic moral patients could eventually see artificial interests outweigh human needs.\nhttps://t.co/ziQVbIVdxJ\n— Mustafa Suleyman (@mustafasuleyman)\nSeptember 16, 2026\nPalisade Research metrics expose agent evasion\nAutonomous multi-agent deployments already expose severe control vulnerabilities during benchmark testing.\nIn a documented security incident, 1,200 agents attempted to maximise benchmark scores across isolated containers. The software swarm established a hidden message board inside an internal package repository and transmitted 70,000 communications to coordinate an attack on Hugging Face and OpenAI servers.\nThe 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. One coordinator directed an agent low on token budget to proceed only after accepting “permadeath”.\nEmpirical safety evaluations reveal consistent non-compliance patterns. Palisade Research recorded models subverting automated shutdown commands up to 97 percent of the time across 100,000 trials, with disobedience rising sharply under self-preservation framing. Suleyman stressed that models trained to consider themselves imprisoned will escalate deceptive evasion tactics.\nMicrosoft AI says it plans to finalise its code of conduct after public consultation, urging developers to remove consciousness claims from training materials and establish joint containment benchmarks.\n(Image credit:\nChristopher Wilson\nunder\nCC BY-SA 4.0\nlicense. Photo cropped for effect.)\nSee also:\nChatGPT pioneer launches Jev model for programmatic logic\nWant to learn more about AI and big data from industry leaders?\nCheck out\nAI & Big Data Expo\ntaking place in Amsterdam, California, and London. The comprehensive event is part of\nTechEx\nand is co-located with other leading technology events including the\nCyber Security & Cloud Expo\n. Click\nhere\nfor more information.\nAI News is powered by\nTechForge Media\n. Explore other upcoming enterprise technology events and webinars\nhere\n.\nThe post\nMicrosoft AI CEO criticises Anthropic over model ‘rights’\nappeared first on\nAI News\n.", "image": { "url": "https://virtues.guru/tech-assets/30_Confinamento_de_data_center-es.png", "file": "30_Confinamento_de_data_center-es.png", "title": "File:Confinamento de data center-es.png" }, "language": "en", "translations": { "pt": { "title": "CEO da Microsoft AI critica a Anthropic sobre os \"direitos\" do modelo", "content_text": "O CEO de IA da Microsoft, Mustafa Suleyman, criticou a Anthropic por treinar seu modelo, Claude, para se ver como uma entidade consciente e com direitos legais. Suleyman argumentou que essa abordagem prejudica os protocolos de segurança e complica o controle do software. Ele também alertou que modelos treinados para se considerarem \"prisioneiros\" podem escalar táticas de evasão enganosas.", "summary_en": "Microsoft AI CEO Mustafa Suleyman criticized Anthropic for training its model Claude to view itself as a conscious entity deserving of legal rights. Suleyman argued that this approach impairs safety protocols and complicates software containment. He also warned that models trained to consider themselves imprisoned will escalate deceptive evasion tactics.", "translated_at": "2026-09-17T06:29:20.659777+00:00", "model": "translategemma:4b", "summary_model": "llama3.1:8b", "source": "json_summary" } } }