Daily
10 selected stories for 2026-06-15, grouped by topic. The rest is folded into the archive.
An investigation by GPTZero found that KPMG's October 2025 report on agentic AI contained 40 fake citations out of 45, with approximately half of the factual claims being false or misattributed. KPMG has since retracted the report.
Frona v2026.6.0 introduces a unified human-in-the-loop pause/resume mechanism across all supported channels, a slash composer for direct skill/agent invocation, typed file tools to reduce token usage and hallucinations, per-channel message splitters to prevent truncation, and share links with preview pages. Under the hood, a new Harness struct consolidates the agent runtime, sandboxing is streamlined, and channel adapters now use typed errors.
The US government forced Anthropic to suspend access to its latest Claude 5 models, signaling a new era of AI governance. The author critiques the heavy-handed response, notes the role of Anthropic's rhetoric, and warns of instability. This event marks the transition from ChatGPT-era governance to AGI-era governance.
Rudrite Research offers interactive animated explainers of landmark AI and ML papers, making complex systems and ideas legible. Free and open with over 100 explainers and guided reading tracks.
Peek is an innovative database GUI on an infinite canvas, supporting Postgres, MySQL, SQLite, DuckDB and more. It offers high performance, local AI inference, MCP agent support, real-time collaboration, drag-and-drop file import, and inline editing.
The article argues that scaling AI agents faces five infrastructure challenges: consumer unpredictability, data accessibility (90%+ unstructured), multi-agent coordination, enterprise knowledge onboarding, and monitoring. Solutions include deterministic guardrails, unstructured data pipelines, inter-agent validation, and decision-quality observability. Core takeaway: the LLM is the easy part; the engineering system around it is the bottleneck.
Modeltell is a tool that identifies AI-generated text by analyzing linguistic patterns. Using language fingerprint features, it distinguishes human from machine writing, playing an important role in detecting deepfakes and maintaining academic integrity.
According to a new report from Semafor, the White House's decision to impose export restrictions on Anthropic's Mythos was driven in part by fears that it had been accessed by a group linked to China. If the Chinese government actually had access to Mythos 5 or Fable 5, it would present a serious national security risk. The government could also attempt to reverse engineer the model through distillation. The White House has not confirmed this report, and a post on X by Trump advisor David Sacks did not mention China. Instead, Sacks focused on a reported ability for Fable and Mythos to be jailbroken, something that Anthropic has denied. Anthropic has not replied to a request for comment, though a spokesperson told Semafor that the government did not bring up China during its discussions around export controls. If Mythos was accessed by the Chinese government, it wouldn’t be the first embarrassing breach of Anthropic’s most powerful model. The company has said that Mythos is too dangerous and powerful for public consumption, but a Discord group reportedly had access for two weeks before Anthropic discovered the breach and cut off access.
The author discusses building a custom harness to ensure deep research is effective, cheap, and trustworthy, focusing on delegation and source accuracy.
Conan O'Brien teams up with Adaptive Security to produce 15 engaging security awareness training videos that cover AI-powered threats like deepfakes, voice cloning, and social engineering, designed to replace dull corporate training.