💪 AI's power struggle | Axios AI+ Summary

1. Key Themes

Theme 1: AI Governance Has No Rulebook — and That's a Crisis

The U.S. government's forced takedown of Anthropic's Fable 5 and Mythos models exposed that frontier AI is currently regulated through emergency improvisation rather than established process. Anthropic was given just 90 minutes to take down its models after Amazon flagged a jailbreak vulnerability to U.S. officials.

"Things feel very ad hoc... you're only as safe as the least safe model on the market." — Adam Gleave, CEO, FAR.AI

"Model oversight has implications for national security, competition and free expression. And right now these decisions are being made through improvised channels, rather than a clear AI regulatory framework."

Theme 2: The Multi-Model Enterprise Strategy Is Becoming Standard

Microsoft's shift to usage-based pricing for Copilot Cowork — and its exploration of a fine-tuned DeepSeek V4 as a cheaper model alternative — signals that enterprise AI platforms are moving away from single-vendor model dependency toward cost-tiered, multi-model architectures.

"The testing reflects Microsoft's broader push toward a multi-model approach, rather than relying only on models from OpenAI and Anthropic."

"We have users who do hundreds of tasks a week, which is great — they're way productive — but the consequence is the costs can go very high." — Charles Lamanna, Microsoft EVP

Theme 3: Agentic AI Creates Runaway Cost Structures

Agentic tools like Copilot Cowork, Claude Code, and OpenAI's Codex compound compute costs by continuously calling models throughout task completion. This is forcing a fundamental rethink of pricing models across the enterprise AI stack.

"Agentic tools like Copilot Cowork, Anthropic's Claude Code and OpenAI's Codex can keep calling AI models as they work through tasks — boosting productivity but also creating bonkers AI bills."

"Microsoft says companies using Copilot Cowork will pay based on how much compute they use."

Theme 4: Foundation Model Bias Has Downstream Consequences at Scale

A small number of foundation models from OpenAI, Google, Meta, and Anthropic power a vast app ecosystem. Bias baked into these models propagates at scale — with documented harms already occurring, such as Meta's Llama 4 recommending conversion "therapy."

"GLAAD is trying to push AI companies before bad practices harden — especially because a small group of foundation models from OpenAI, Google, Meta and Anthropic now feeds a much larger app ecosystem."

"The report cites harms including Meta's updated Llama 4 model recommending conversion 'therapy' in response to queries and moderation systems that wrongly suppress legitimate LGBTQ content."

Theme 5: Niche AI Consumer Apps Are Finding Funding in the Void Left by Restricted Free Tiers

As free AI video generation options become more constrained, specialized consumer applications are carving out investable market positions. Nauk Nauk's $20M raise targeting the "kidult" market is an early signal.

"Nauk Nauk is launching as free AI video-generation options become more limited, creating an opening for specialized consumer apps."

2. Contrarian Perspectives

Perspective 1: Government AI Intervention May Be More Dangerous Than Corporate Self-Regulation

The conventional view is that corporate self-policing is insufficient for frontier AI oversight. But the Fable/Mythos takedown illustrates a less-discussed risk: government decisions made through informal, opaque channels (e.g., routing concerns through the Treasury Secretary) may be equally — or more — dangerous than industry self-governance.

"When Amazon flagged its concerns to the administration, it went to Treasury Secretary Scott Bessent, who was already close to the Mythos rollout because of its planned use by major U.S. banks."

"Handing the reins to companies means trusting firms with huge commercial incentives. Leaving decisions up to governments could lead to secretive high-stakes decisions with little public approval."

The proposed solution — a new AI-fluent agency combining company testing, outside audits, and government authority — implicitly acknowledges neither existing option is adequate.

Perspective 2: AI Job-Loss Doom May Be Investor-Motivated Narrative, Not Reality

The mainstream narrative is that AI will cause a net loss of jobs. Microsoft's Brad Smith pushes back, suggesting the alarm may be more about investment positioning than technology truth.

Brad Smith "said 'let's not panic' about AI's alleged hit to the labor market, signaling that the doom and gloom over AI and work may be related to investment goals rather than truth about the technology."

Perspective 3: DeepSeek — a Chinese AI Model — Could Become Core Enterprise Infrastructure in the U.S.

Conventional wisdom treats Chinese AI models as a national security liability for U.S. enterprises. Microsoft's active consideration of DeepSeek V4 for Copilot Cowork challenges this, with the company arguing that cloud hosting on Azure with enterprise security controls neutralizes the risk.

"If Microsoft goes forward with DeepSeek, the company says, the model would be optional for customers and fully hosted on Azure, keeping customer data within Microsoft's cloud and covered by Azure's enterprise security, compliance and data-residency controls."

3. Companies Identified

Anthropic

Description: AI safety-focused frontier model company
Why mentioned: Central case study in AI governance failure — its Fable 5 and Mythos models were forced offline by U.S. government intervention following a reported jailbreak
Quote: "Anthropic was given 90 minutes last week to take down Fable and Mythos after Amazon raised concerns with U.S. officials about a jailbreak that could bypass Fable's guardrails and expose cybersecurity capabilities."

Microsoft

Description: Enterprise technology giant; maker of Copilot suite
Why mentioned: Shifting Copilot Cowork to usage-based pricing; considering DeepSeek V4 as a lower-cost model option; emblematic of multi-model enterprise strategy
Quote: "Microsoft is moving Copilot Cowork to usage-based pricing as it expands access to the enterprise AI tool — and is considering a Microsoft-hosted version of DeepSeek as a cheaper model option."

DeepSeek

Description: Chinese open-source AI model developer
Why mentioned: Being evaluated by Microsoft as a cost-effective model alternative for enterprise use, raising geopolitical and competitive questions
Quote: "The company tells Axios it is exploring a fine-tuned version of DeepSeek V4, or another open-source model, as a lower-cost alternative to the Anthropic and OpenAI models now powering Copilot Cowork."

FAR.AI

Description: AI safety research firm
Why mentioned: CEO cited as a key voice on the ad hoc nature of current AI governance
Quote: "Things feel very ad hoc... you're only as safe as the least safe model on the market." — Adam Gleave, CEO

ControlAI

Description: AI governance advocacy organization
Why mentioned: U.S. executive director cited arguing for government authority over AI regulation, distinct from industry self-regulation
Quote: "In every sport, you have to separate the referee from the team players... regulation should never come from industry." — Connor Leahy

Nauk Nauk

Description: AI video startup; turns toy photos into short animated videos
Why mentioned: Raised $20M; targeting "kidult" market; positioned as a specialized consumer app in a space where free AI video tools are retreating
Quote: "Nauk Nauk is launching as free AI video-generation options become more limited, creating an opening for specialized consumer apps."

Meta

Description: Social media and AI company; maker of Llama models
Why mentioned: Cited as a negative example — its Llama 4 model recommended conversion "therapy" in response to LGBTQ queries
Quote: "The report cites harms including Meta's updated Llama 4 model recommending conversion 'therapy' in response to queries."

GLAAD

Description: LGBTQ media advocacy organization
Why mentioned: Released a new AI bias report ("Build for Everyone") calling out harmful practices baked into foundation models and offering industry recommendations
Quote: "Neutrality is no longer an option." — GLAAD CEO Sarah Kate Ellis

SpaceX

Description: Aerospace and technology company founded by Elon Musk
Why mentioned: Briefly noted as surpassing Amazon in market value less than a week after going public, becoming the fifth largest U.S. company
Quote: "SpaceX pushed past Amazon in value, becoming the fifth largest U.S. company less than a week after going public."

4. People Identified

Adam Gleave

Description: CEO, FAR.AI (AI safety research firm)
Why mentioned: Critiqued the ad hoc nature of current U.S. AI governance
Quote: "Things feel very ad hoc... you're only as safe as the least safe model on the market."

Connor Leahy

Description: U.S. Executive Director, ControlAI
Why mentioned: Argued that AI regulation authority must sit with government, not industry
Quote: "In every sport, you have to separate the referee from the team players... regulation should never come from industry."

Charles Lamanna

Description: Executive Vice President for Copilot, Agents and Platform, Microsoft
Why mentioned: Explained Microsoft's rationale for shifting to usage-based pricing due to spiraling agentic AI compute costs
Quote: "We have users who do hundreds of tasks a week, which is great — they're way productive — but the consequence is the costs can go very high."

Sarah Kate Ellis

Description: CEO, GLAAD
Why mentioned: Authored/presented findings from GLAAD's new AI bias framework report
Quote: "Neutrality is no longer an option."

Brad Smith

Description: Vice Chair and President, Microsoft
Why mentioned: Pushed back against AI job-loss doom narrative, suggesting it may be tied to investment motivations
Quote: Said "let's not panic" about AI's alleged hit to the labor market, "signaling that the doom and gloom over AI and work may be related to investment goals rather than truth about the technology."

Scott Bessent

Description: U.S. Treasury Secretary
Why mentioned: Served as the informal channel through which Amazon's AI safety concerns were routed to the U.S. government — illustrating the absence of a formal regulatory process
Quote: "When Amazon flagged its concerns to the administration, it went to Treasury Secretary Scott Bessent, who was already close to the Mythos rollout because of its planned use by major U.S. banks."

Daniel Liu

Description: Co-founder, Nauk Nauk
Why mentioned: Disclosed the company's $20M raise and its market positioning
Quote: Nauk Nauk "raised $20 million to expand its AI video app that turns photos of toys into short animated videos."

5. Operating Insights

1. Build usage-based pricing into agentic AI products from day one — unlimited plans are economically unviable.

Microsoft discovered through live deployment that heavy agentic users drive costs to unsustainable levels under flat-rate pricing. Operators building agentic tools should architect consumption-based billing from the start rather than retrofitting it.

"Charles Lamanna... told Axios that testing showed Copilot Cowork could not be offered on an unlimited-use basis. 'We have users who do hundreds of tasks a week, which is great — they're way productive — but the consequence is the costs can go very high.'"

2. Enterprises should audit foundation model choices for embedded bias before deployment — not after.

GLAAD's report demonstrates that harmful outputs (e.g., conversion therapy recommendations from Llama 4) are already occurring in production. Operators deploying third-party foundation models inherit those models' biases; proactive auditing is now a compliance and reputational necessity.

"The problems GLAAD flags — biased training data, privacy risks, automated discrimination, misinformation and the suppression of legitimate speech — extend beyond LGBTQ users to other minorities and groups in political disfavor."

3. When deploying politically sensitive or open-source models (including foreign-origin models like DeepSeek), cloud isolation and fine-tuning are the enterprise security playbook.

Microsoft's framework for potentially deploying DeepSeek — Azure hosting, enterprise security controls, fine-tuning, and added safeguards — offers a replicable template for enterprises wanting cost efficiency without geopolitical or data exposure risk.

"The model would be optional for customers and fully hosted on Azure, keeping customer data within Microsoft's cloud and covered by Azure's enterprise security, compliance and data-residency controls. Microsoft says it has also fine-tuned the model and added safeguards, including changes aimed at reducing bias."

6. Overlooked Insights

1. The Mythos model had already been scoped for deployment by major U.S. banks — raising significant questions about financial sector AI risk.

The article briefly notes that Treasury Secretary Bessent was involved because Mythos was planned for use by "major U.S. banks." This suggests frontier AI models with unresolved security vulnerabilities were on the verge of being embedded into systemically important financial infrastructure — a risk that received no dedicated attention in the piece.

"Treasury Secretary Scott Bessent... was already close to the Mythos rollout because of its planned use by major U.S. banks."

2. The proposed solution to AI governance — a new AI-fluent agency — represents a significant potential market for regulatory technology and compliance infrastructure.

The article floats the idea of "an entirely new AI-fluent agency tasked with combining company testing, outside audits and government authority." If such an agency materializes, it would create demand for auditing tools, compliance frameworks, and third-party testing infrastructure — an emerging market opportunity that goes unexamined.

"The most workable answer may be neither company self-policing nor unilateral government control, but an entirely new AI-fluent agency tasked with combining company testing, outside audits and government authority."