What the Pope Actually Said About AI
The AI Daily Brief covers Anthropic's Mythos model rollout under Project Glasswing (finding 10,000+ critical vulnerabilities), DeepSeek's permanent price cuts and $10B funding round, Grok V9 Medium's training completion, and a deep analysis of Pope Leo XIV's first encyclical 'Magnifica Humanitas' on AI and human dignity.
Summary
The episode opens with an update on Project Glasswing, Anthropic's limited rollout of their Mythos model to ~50 partners focused on cybersecurity. The project has identified over 10,000 high or critical severity software vulnerabilities, with partners like Mozilla fixing 271 bugs — more than 10x what they found using the previous Opus 4.6 model. Critically, the bottleneck has shifted from finding vulnerabilities to patching them, as human capacity to triage and fix bugs hasn't kept pace with AI's ability to discover them. Box CEO Aaron Levy dubbed this 'Jevons Paradox' in action, predicting a security engineer boom rather than job elimination. Globally, European, UK, Canadian, Australian, and Asian governments have pressured Anthropic for access to Mythos, while Anthropic maintains that no company has yet developed sufficient safeguards for a public release.
A separate headline reveals the White House approved a secret $9 billion budget for the CIA and NSA to build their own AI inference clusters using NVIDIA Blackwell chips. The intelligence agencies reportedly fell behind in AI build-out and need to run Mythos-class models in classified settings. Notably, the NSA and CIA are reportedly more comfortable than the Pentagon with Anthropic's restrictions on autonomous weaponry and domestic surveillance, and a contract between Anthropic and the NSA is reportedly being finalized.
On the competitive landscape, DeepSeek made its 75% token discount permanent, pricing its V4 model at $0.44/million input tokens — far below OpenAI and Anthropic offerings — while announcing a $10 billion funding round valuing the company at $45 billion, with participation from the Chinese government's AI fund and companies like Tencent. Meanwhile, XAI completed training on Grok V9 Medium, a 1.5 trillion parameter model expected to release in 2-3 weeks, with GrokBuild also expanding to more subscription tiers.
The main episode focuses on Pope Leo XIV's encyclical 'Magnifica Humanitas,' subtitled 'On Safeguarding the Human Person in the Time of Artificial Intelligence.' The host argues that most social media reactions misrepresented the document, either cherry-picking quotes to support pre-existing views or criticizing it for not engaging with their preferred frameworks. The host characterizes the encyclical's central argument as a defense of human dignity that cannot be reduced to an intelligence benchmark — even if AI surpasses humans cognitively, it remains categorically different because it lacks embodiment, experience, emotion, moral conscience, and spiritual growth. The host highlights paragraph 99, where Leo explicitly rejects equating AI intelligence with human intelligence, and paragraph 118, where Leo argues that human limitations like illness and suffering are not defects to be fixed but conditions through which humanity matures. The host frames the encyclical as a foundational, contra-transhumanist document that plants a flag ahead of increasingly heated debates about AI cognition and personhood.
Key Insights
- Anthropic's Mythos model has shifted the cybersecurity bottleneck from vulnerability discovery to patching — partners are finding bugs 10x faster than before, but human capacity to triage and fix them hasn't scaled, leading some maintainers to ask Anthropic to slow down disclosures.
- The host argues that Pope Leo XIV's encyclical is not a condemnation of AI but a foundational, flag-planting document that positions humanity at the beginning of a long debate, with its core thesis being that human value cannot be reduced to an intelligence benchmark regardless of AI capability.
- The host contends that most social media reactions to the encyclical were poor — critics on both sides cherry-picked quotes, with some angry it didn't condemn AI exploitation strongly enough and others frustrated it didn't engage with AI cognition or personhood theories.
- DeepSeek's decision to make its 75% discount permanent, combined with a $45B valuation funding round backed by the Chinese government, signals a deliberate strategy to treat AI tokens as a commodity and capture global developer share, especially as US model prices rise.
- The Pentagon and intelligence agencies diverge on AI contract terms: the Pentagon demanded AI be usable for 'any lawful use' including autonomous weapons, while the NSA and CIA — legally barred from domestic surveillance — are reportedly more comfortable with Anthropic's restrictions, potentially unlocking a path to broader government AI contracting.
Topics
Full transcript available for MurmurCast members
Sign Up to Access