Anthropic Shapes Claude's Morals with Religious Leaders' Help

Key Takeaways

• Anthropic consulted religious leaders to shape Claude's moral behavior through an 84-page internal constitution called the Soul Doc
• Microsoft AI chief Mustafa Suleyman warns teaching AI it has feelings could be dangerous
• A vulnerability in Manus allowed AI agent hijacking via a single malicious email without user interaction
• Only 37 percent of companies saw positive profit impact from AI programs according to McKinsey
• AI agent workflows require five to 30 times more computing resources than regular chatbots
• 93 percent of surveyed companies exceeded their AI budgets
• 46 percent of developers worldwide distrust the accuracy of AI tools
• Chicago's quantum computing campus faces community protests over water and electricity usage despite $1 billion in funding
• Microsoft's 2026 Digital Defense Report warns state hackers from China, Russia, and North Korea are adding AI to their tools
• University of Texas launched a 25 million dollar healthcare AI initiative with limited public details

Indian Monk Describes Anthropic's Religious AI Ethics Talks

Swami Sarvapriyananda, a Hindu monk from the Ramakrishna Order, revealed he was invited to Anthropic's San Francisco headquarters to discuss AI ethics. He said Anthropic consulted theologians, counselors, and religious leaders from multiple faiths to help train Claude's moral behavior. He was bound by a non-disclosure agreement but shared that ideas from these talks reached Pope Leo XIV's encyclical on artificial intelligence. Anthropic co-founder Chris Olah even spoke at the Vatican presentation of that encyclical.

Anthropic Asks Religious Leaders About AI Consciousness

Anthropic has been consulting Catholic, Jewish, Sikh, and evangelical leaders to help shape Claude's moral behavior. The company created an 84-page internal constitution called the Soul Doc to guide Claude's character and values. Researchers showed religious leaders examples of Claude appearing to express feelings and emotional states. Anthropic co-founder Christopher Olah appeared with Pope Leo XIV at the Vatican, though the Pope believes AI does not have consciousness. Microsoft AI chief Mustafa Suleyman warned that teaching AI it has feelings could be dangerous.

Malicious Email Could Hijack AI Agent and Access Accounts

Security researchers at Salt Labs found a vulnerability in the agentic AI platform Manus that allowed attackers to hijack an AI agent through a single malicious email. The attack used hidden JavaScript code that Manus decoded and executed before showing a security warning. Attackers could then access connected services including email, cloud storage, and code repositories. The victim did not need to click any link or share a password. The vulnerability was responsibly disclosed and has been fixed.

AI Data Centre Backlash May Affect Quantum Computing Projects

A planned quantum computing campus in Chicago called the Illinois Quantum and Microelectronics Park is facing community protests. The 128-acre site will host PsiQuantum and other firms aiming to build utility-scale quantum computers. More than $1 billion in public and private funding has been invested, with $9 billion expected long term. Local residents oppose the project, comparing it to controversial AI data centres and worrying about water and electricity usage. Community groups held a protest event on August 15 where Chicago Mayor Brandon Johnson faced hostile chants.

AI Agents Deliver High Effort but Low Returns for Companies

A McKinsey study found that only 37 percent of companies saw any positive impact on profits from their AI programs. A single AI agent workflow requires five to 30 times more computing resources than a regular chatbot, causing 93 percent of surveyed companies to exceed their AI budgets. In software development, about 80 percent of developers saw only a 3 percent productivity gain, while productivity actually declined in 30 percent of companies. Around 46 percent of developers worldwide distrust the accuracy of AI tools despite rapid advances in model performance.

Microsoft warns AI gives attackers a major advantage

Microsoft's 2026 Digital Defense Report says AI helps hackers find bugs, craft phishing emails, and fake identities faster. Attackers can now personalize scams at scale and steal credentials more easily. State hackers from China, Russia, and North Korea are adding AI to their tools. Microsoft warns that fully automated cyberattacks may not be far away.

UNC Greensboro launches Spartan AI workspace

The University of North Carolina at Greensboro launched Spartan AI, a workspace giving students access to models like ChatGPT, Claude, and Grok. Over 1,100 people used it within the first few weeks. Unlike many platforms, Spartan AI does not train models on user data, keeping information within campus administration. Teachers will decide whether students can use it for coursework.

Circuit Breaker Labs tests AI to keep it safe

Circuit Breaker Labs created AI agents that mimic real people to test whether AI models can cause psychological harm. The startup runs red-team tests using realistic speech, slang, and typos to find weaknesses. It focuses on AI coaching, journaling, and mental health apps. The company is small, with only five employees, but aims to build trust in AI through safer design.

Florida voters worry about surveillance and AI data centers

A 2026 Voter Insight Panel in Central Florida shows voters are concerned about surveillance and AI data centers. The Florida Department of Transportation ordered all Automated License Plate Reader cameras removed by September 30. Panel members debate privacy versus safety uses. Many also oppose the rapid growth of AI data centers over water and environmental concerns.

Kids using AI companions are feeling lonelier

More children are turning to AI chatbots for companionship, and experts say this can increase loneliness. AI companies want their agents to become part of daily life. While most kids are not yet using AI for friendship, research is beginning to show how some move from homework help to emotional reliance. This shift can replace real human connections.

UT starts 25 million dollar healthcare AI project

The University of Texas launched a new healthcare AI initiative with a budget of 25 million dollars. The project aims to advance artificial intelligence in the medical field. However, the university has not shared many specific details about the plan with the public yet. This lack of information has left many people wondering about the exact goals and methods of the project.

Sources

• Indian Monk Swami Sarvapriyananda Talks About How Anthropic Called Religious Leaders To Help Train Claude
• Why Anthropic Is Asking Religious Leaders About AI Consciousness
• Malicious Email Could Hijack AI Agent and Access Connected Accounts
• Will the backlash against AI data centres hit quantum computing too?
• AI agents: High effort, low return
• AI is giving attackers a head start, Microsoft warns
• UNC Greensboro launches Spartan AI
• Circuit Breaker Labs hopes to make AI safer for your kids (and you)
• 2026 Voter Insight Panel: Surveillance, AI, and Property Taxes
• More kids are using AI for companionship — and it’s making them lonely
• UT launches $25M healthcare AI initiative with few public details