👀 Claude complaints | Axios AI+ Summary

1. Key Themes

AI Access Is Stratifying: Frontier Capability Is Becoming Pay-to-Play

The most powerful AI models are increasingly being gated behind premium pricing tiers, usage-based billing, and exclusive enterprise programs. This creates a growing gap between power users and casual users.

"Advanced capabilities are increasingly gated behind higher-cost tiers, API usage or experimental programs... Anthropic recently moved large enterprise customers to a fully usage-based (token) pricing model, tying intelligence more directly to spend."

Default AI Experiences May Deteriorate Even as Frontier Models Improve

As labs concentrate compute on their most advanced systems, mainstream product quality may quietly erode — creating a structural tension between innovation at the top and regression at the base.

"What we're watching: Whether 'default' AI experiences continue to get worse even as frontier systems get dramatically stronger."

AI-Powered Cybersecurity Is a National Security Priority for Financial Infrastructure

Government officials and major financial institutions are treating AI cyber capability as a systemic risk requiring coordinated early access and real-world testing — not just a product feature.

"The access comes after Treasury Secretary Scott Bessent and Fed Chair Jerome Powell called a meeting with the biggest names on Wall Street to discuss Mythos... The meeting focused on risks of AI-powered attacks on bank systems as well as preventative measures."

AI in Life Sciences Is a Policy Battle, Not Just a Technology Race

OpenAI is actively lobbying to reshape the regulatory and data-access environment for AI in biomedical research, framing it as a national resource question. Infrastructure — compute, labs, energy — is as critical as the models themselves.

"It calls for greater access to medical and scientific data, treating advanced AI as a national research resource, and pushes for investment in the 'physical stack,' meaning compute, labs, energy and infrastructure."

Enterprise AI Adoption Is Accelerating in Legacy Institutions

Traditional financial incumbents are not just experimenting with AI — they are deploying it at scale with dedicated digital workers and achieving outsized market performance as a result.

"The bank... has over 100 digital employees that have their own tasks, managers and email addresses. Under Vince's leadership, BNY rose to the best-performing stock in an index tracking a group of major banks, up 218%."

2. Contrarian Perspectives

The "Nerfed AI" Narrative Is Likely Exaggerated — But the Perception Problem Is Real and Damaging

Users widely believe Anthropic deliberately degraded Claude to redirect compute toward frontier models. Even Claude itself, when asked directly, pushed back on the most extreme version of this theory — yet the perception still carries serious commercial risk.

"Anthropic made real configuration changes that objectively reduced default thinking depth across all surfaces including claude.ai, but the most extreme 'secret nerfing' narrative overstates what happened," — Claude, when asked by analyst Patrick Moorhead.

The more uncomfortable explanation: users may simply have habituated to what once felt magical, raising their baseline expectations over time.

"Another theory is that users aren't seeing decline so much as acclimating to what previously felt magical. Over time, expectations rise and flaws become more noticeable — a phenomenon known as habituation."

AI Drug Discovery Hype Remains Far Ahead of Clinical Reality

Despite enormous enthusiasm for AI in biomedical research, actual deployments have barely crossed the threshold into human trials — a sobering fact buried inside OpenAI's own optimistic policy report.

"Only a few AI-discovered or AI-designed drugs have reached clinical trials."

The "Eroom's Law" framing also cuts against the grain of Silicon Valley optimism: scientific productivity slows as knowledge accumulates, and AI may face the same compounding complexity problem it claims to solve.

"'Eroom's Law' — Moore's Law in reverse — holds that scientific research slows as knowledge accumulates, because each new generation of scientists must spend more time just absorbing what came before."

OpenAI's Life Sciences Report Is a Lobbying Document, Not a Research Paper

Framed as scientific advocacy, the report is primarily a policy pitch designed to reshape data access rules and federal investment priorities in OpenAI's favor — investors should read it as such.

"OpenAI's report functions as a policy pitch, arguing for changes that would support broader AI use in life sciences."

3. Companies Identified

Anthropic

Description: AI safety-focused lab, maker of Claude
Why mentioned: Central to both the Claude performance backlash story and the BNY cyber model access story; testing advanced model Mythos
Quote: "Anthropic says it adjusted the default level of reasoning in Claude Code, but denies the changes were tied to compute constraints or Mythos."

OpenAI

Description: Leading AI lab, maker of GPT series
Why mentioned: BNY has early access to GPT-5.4-Cyber; published a life sciences lobbying report; taking a broader distribution approach to cyber models than Anthropic
Quote: "OpenAI is making tools more widely available for defensive cyber work while still preventing nefarious actors from accessing them."

BNY (Bank of New York Mellon)

Description: America's oldest bank; aggressive AI adopter
Why mentioned: Featured as a case study in enterprise AI deployment; has early access to both OpenAI and Anthropic's advanced cyber models; 218% stock performance under current CEO
Quote: "BNY, America's oldest bank, has early access to OpenAI's and Anthropic's advanced cyber capability models... making the bank one of few vetted enterprises with early access."

Allbirds

Description: Sneaker company pivoting to AI
Why mentioned: Brief but notable mention of a 580% stock pop after announcing an AI pivot with $50 million in funding — a signal of market sentiment around AI pivots
Quote: "Thinking about the 580% stock pop for Allbirds after the sneaker company decided to become an AI firm with just $50 million in funding."

ElevenLabs / Speechify

Description: AI voice generation companies
Why mentioned: Flagged by Sen. Maggie Hassan in letters probing how AI voice firms are preventing scam exploitation
Quote: "Sen. Maggie Hassan (D-N.H.) sent letters to ElevenLabs, Speechify and other AI voice firms for details on how they're stopping scammers."

Apple

Description: Consumer tech giant
Why mentioned: Its privacy-first reputation is identified as a potential competitive differentiator in the AI era
Quote: "Apple's reputation for enhanced privacy could become a selling point in the AI era."

Arm

Description: Semiconductor IP company (newsletter sponsor)
Why mentioned: Sponsor context; positioning itself for the agentic AI infrastructure era with the Arm AGI CPU
Quote: "Arm is enabling this shift with the Arm AGI CPU, built for real-time orchestration at scale."

4. People Identified

Robin Vince

Description: CEO of BNY
Why mentioned: Directly confirmed early access to OpenAI and Anthropic cyber models; credited with BNY's 218% stock outperformance; leading BNY's aggressive AI transformation
Quote: "Anthropic and OpenAI recognize the importance of releasing their cyber-capable models to certain institutions early... 'and in our case, obviously the financial services world,' Vince says."

Boris Cherny

Description: Head of Claude Code at Anthropic
Why mentioned: Anthropic pointed to his March 6 post on X as the official explanation for Claude's changed behavior — framing the controversy as a configurable default, not a secret downgrade
Quote: "You can change it anytime in the /model selector if you prefer low effort (faster) or high effort (more intelligence). The setting is sticky and will persist for your next session."

Patrick Moorhead

Description: Technology analyst
Why mentioned: Took the creative step of asking Claude itself to adjudicate the "nerfing" controversy, surfacing a useful on-the-record response from the model
Quote: "Analyst Patrick Moorhead decided to ask Claude to weigh in."

Scott Bessent

Description: U.S. Treasury Secretary
Why mentioned: Co-convened an urgent meeting with Wall Street CEOs to discuss the cybersecurity risks of Anthropic's Mythos model
Quote: "Treasury Secretary Scott Bessent and Fed Chair Jerome Powell called a meeting with the biggest names on Wall Street to discuss Mythos."

Jerome Powell

Description: Federal Reserve Chair
Why mentioned: Jointly convened the Wall Street cybersecurity briefing on Mythos alongside Bessent
Quote: "Treasury Secretary Scott Bessent and Fed Chair Jerome Powell called a meeting with the biggest names on Wall Street to discuss Mythos."

Sen. Maggie Hassan

Description: U.S. Senator (D-N.H.)
Why mentioned: Taking legislative action on AI voice scam risks, sending formal letters to ElevenLabs, Speechify, and other AI voice companies
Quote: "Sen. Maggie Hassan (D-N.H.) sent letters to ElevenLabs, Speechify and other AI voice firms for details on how they're stopping scammers."

5. Operating Insights

Power Users Should Manually Override Default AI Settings, Not Accept Them Passively

The Claude controversy reveals that default model settings — now optimized for cost efficiency or speed — may no longer reflect peak capability. Operators building on AI platforms should audit and explicitly configure reasoning depth and model tiers for their specific workflows rather than assuming defaults are optimal.

"You can change it anytime in the /model selector if you prefer low effort (faster) or high effort (more intelligence). The setting is sticky and will persist for your next session." — Boris Cherny, Head of Claude Code

Seek Early Enterprise Access Programs for Frontier AI Models — They Deliver Competitive Intel and Feedback Leverage

BNY's early access to both OpenAI and Anthropic's most sensitive cyber models was not accidental. Labs want vetted real-world testers, and being in that cohort provides both capability advantages and direct influence on model development.

"The AI labs also want feedback and real-world testing... Other firms with access to these previews will be able to share lessons learned with one another as well as the labs themselves."

6. Overlooked Insights

OpenAI and Anthropic Are Pursuing Meaningfully Different Distribution Strategies for Dangerous AI Capabilities

Anthropic restricted Mythos to roughly 40 enterprises. OpenAI is rolling GPT-5.4-Cyber out more broadly for defensive use. This divergence in openness philosophies — not just model capability — may become a key competitive differentiator for enterprise customers and a fault line for regulators to exploit.

"OpenAI's new model variant, GPT-5.4-Cyber, will be rolled out to a broader set of organizations than Anthropic's Mythos, which initially reached about 40 enterprises. While Anthropic signaled that its model was too dangerous to release broadly, OpenAI is making tools more widely available for defensive cyber work."

Government Cybersecurity Cuts Are Undermining the U.S.'s Ability to Respond to Its Own AI Risk Warnings

While top officials are urgently briefing Wall Street on AI-powered cyber threats, domestic cybersecurity infrastructure is simultaneously being reduced — creating a direct contradiction between threat awareness and defensive capacity.

"Government cybersecurity cuts and its Anthropic lawsuit are complicating the Trump administration's response to Anthropic's new Mythos model."