AI-Hallucinated Citations Infiltrate Medical Papers — Clinical Guidelines Affected Too
A research team including Columbia University audited 2.5 million biomedical papers and found that fabricated references believed to be AI-generated have increased more than twelvefold since 2023. The problem: some of these papers have actually been used to shape real clinical guidelines.
While AI's text generation capabilities can help streamline research, they do not guarantee citation accuracy. The possibility of fabricated citations influencing clinical decision-making amounts to a public health risk that goes beyond mere academic misconduct. The research team points to a link with the proliferation of language models, making stronger human verification processes indispensable for anyone using AI tools.
Altman Revises His "End of Jobs" Claim
OpenAI's Sam Altman has revised his earlier stance on the possibility of massive AI-driven job destruction. According to Reuters, Altman said that AI is unlikely to bring about the end of jobs.
It is noteworthy when the head of the AI industry publicly corrects his own past predictions. That said, as MIT Technology Review's analysis shows, there is also data indicating that, among early-career workers one to three years into their jobs, employment in roles with high AI exposure has fallen 16% in relative terms. Altman's remarks mean there will be no large-scale employment crisis — it is worth noting they do not deny the impact on specific groups.
ING Builds a Trading System with Vibe Coding
Bloomberg reports that ING, a major Dutch financial institution, is building a new trading system using "vibe coding," an AI-driven code generation approach.
"Vibe coding" refers to a development method in which you give instructions to an AI in natural language and have it generate code. Applying this method to a financial institution's mission-critical systems suggests that enterprise adoption of AI coding is advancing faster than expected. At the same time, it is a case that once again raises questions about how to assure the quality of generated code and how security audits should be handled.
Spotify's CEO Defends AI-Generated Music
Spotify CEO Daniel Ek has defended the presence of AI-generated music on the platform. According to the FT, Ek indicated that while artists' rights must be respected, Spotify has no intention of excluding AI music.
AI's presence in the music industry has grown stronger year by year, and streaming distribution of AI-generated tracks is already becoming commonplace. The CEO of the largest streaming service explicitly embracing AI music is a significant statement about the direction of the industry as a whole. However, the fundamental challenges of copyright and fair compensation for artists remain unresolved.
Block Open-Sources Its Internal AI Agent "Goose"
Block, the company behind Square and Cash App, has released "Goose" — an AI agent adopted by 60% of its employees — as open source.
Goose is an AI development assistant widely used within Block, supporting everything from code generation to review and testing. Its high internal adoption rate demonstrates that AI agents can be deeply embedded in development workflows. With the open-source release, other companies and developers can now build similar AI-assisted development environments of their own.
Rethinking the Limits of LLM Code Generation Through Amdahl's Law
A Hacker News discussion titled "Amdahl's Law for LLM-Generated Code" has been drawing attention. The argument: while LLMs can theoretically generate millions of lines of code, the only way to guarantee the correctness of critical code is manual auditing — and auditing is harder than writing.
This suggests that a fundamental ceiling exists on productivity gains from LLM code generation. As long as the speed at which humans can understand and verify code remains the bottleneck, faster code generation alone will cap overall throughput. How far advances in test automation and formal verification can push this ceiling will be the key question going forward.
Microsoft Releases an AI Agent Governance Toolkit
Microsoft has open-sourced the "Agent Governance Toolkit," a toolkit for the operational management of AI agents. It provides a framework for monitoring AI agent behavior, enforcing access control, and supporting compliance.
As enterprise adoption of AI agents accelerates, the absence of governance becomes a serious risk. By providing a standard tool in this space, Microsoft is expected to help raise the baseline of AI agent safety across the industry.
---