Meta Introduces "AI Mode" on Facebook — A Cross-Platform AI Experience

On June 15, Meta announced that it will roll out a new suite of AI features called "AI Mode" on Facebook. The feature draws on public information from across Meta's family of platforms, including Instagram and Threads, to answer user questions. The move appears aimed at regaining ground in the AI race and keeping engagement within Meta's own platforms.

While this is a natural evolution for Meta, the design — which leverages data across platforms even if that data is public — warrants close scrutiny from a privacy perspective.

US Government Demands an "Unhackable LLM" from Anthropic — An Impossible Order?

US government officials have been criticizing Anthropic, accusing the company of ignoring the Trump administration's cyber directives and releasing its model "Fable 5" without approval. One senior administration official said, "They betrayed us," and discussions with the Department of Commerce, the CIA, and science advisor Michael Kratsios are reportedly underway.

The government is reportedly demanding an "unhackable LLM," an ask that is technically extremely difficult to fulfill. How to reconcile security requirements with the timing of model releases looks set to become a challenge for the entire AI industry.

KPMG Retracts AI-Generated Report — A Major Incident Caused by Hallucination

KPMG, a major accounting firm, has officially retracted an AI-generated report after it was found to contain hallucinations (generated content that contradicts facts). Details are still awaiting further reporting, but the fact that one of the so-called Big Four audit firms published AI-generated output as an official report and had to pull it once problems surfaced underscores the importance of AI governance in the enterprise.

This case shows that fact-checking and human review processes are indispensable when embedding AI into business workflows.

AI-Generated DNA Startup Radical Numerics Raises $50 Million

Fortune has learned that Radical Numerics, a startup developing technology to generate DNA sequences using AI, has raised $50 million (roughly 7.5 billion yen). Led by Eric Nguyen, the company is targeting applications in biodefense and drug discovery.

AI-driven DNA design holds the potential to accelerate drug development and enable biological experiments that were previously impossible. At the same time, biosecurity risk management will be required in parallel with the technology's development.

Local Coding Agents Have Reached Practical Levels — But They Still Need "Supervision"

The LocalLLaMA community on Reddit has seen a lively debate over the practicality of coding agents that run in local environments. One poster said they are "extremely handy for small fixes, reading repositories, file changes, and tedious coding work," while also pointing out that "if you give them too much freedom, they'll touch unrelated files or write broken code that looks correct at first glance."

As a practical workflow, the cycle of "assign a small task → run the tests → review the diff → fix anything that looks off → repeat" reportedly works well. Rather than full autonomy, the best practice at this point seems to be a setup where humans check in periodically, acting as a manager.

---