The most relevant in artificial intelligence, every day.
Adrafinil is an open-source CLI tool for macOS that prevents the computer from sleeping only while agents such as Claude Code, Cursor or Aider are actively running. Unlike caffeinate or Amphetamine, it integrates directly with the agents' hooks and releases the system when it detects inactivity.
Entertainment giant Hasbro is reportedly asking child actors from Peppa Pig to give up their voices for use by artificial intelligence. The move involves licensing the children's voices for AI applications.
The GPT-5.6 Soul Ultra model outperformed Claude Mythos on Stanford's terminals benchmark and beat Claude Fable 5 by a significant margin. A separate model, Terra, also surpassed Fable 5 in cybersecurity tests.
A woman whose town installed AI-powered traffic cameras grew alarmed after realizing the systems may capture more than traffic violations. She called the broader tracking "so much creepier."
Anthropic filed its confidential S-1 draft on June 1, 2026, and OpenAI followed on June 8. Both AI giants, valued near $1 trillion, are aiming to go public in late 2026 or early 2027, with Goldman Sachs and Morgan Stanley as shared advisers.
An opinion piece in Deseret News contrasts two radically opposite uses of companion chatbots: the ElliQ robot that reduces loneliness in the elderly by 95% and the Character.AI chatbot linked to the suicide of a 14-year-old. The key, the author says, lies not in the technology but in whether the design returns the user to human life or replaces it.
An analysis published in Medical Xpress warns about the risks of users and systems equating AI mental health tools with real clinical psychotherapy, a confusion with potentially serious consequences for patients.
Z.ai launches GLM-5.2, a 750-billion-parameter open-source model that comes within just one percentage point of Anthropic's Opus 4.8 on agentic evaluations, but at one-sixth of the cost. Chinese models already hold the top four spots in OpenRouter's global usage ranking.
An AI news roundup groups several topics: an alleged Claude Opus 4.7 leak, a full-stack Anthropic application, a new GPT model and Claude Code updates. There is no verifiable content beyond the headline.
The newsletter mentioned: McKinsey partner says up to 50% of work hours could be transformed within the next five years
A researcher publishes MobileGuard on Zenodo, the first governance framework designed specifically for agentic AI on mobile platforms. With four operational pillars and three empirical studies, it claims to reduce deployment errors by 74.1% against real pipelines on iOS and Android.
A developer releases an offline, dependency-free benchmark that measures the four critical failure modes of AI agent memory systems: retraction, collision, retrieval and conflict. Traditional retrieval metrics hide vast differences in real performance (from 23% to 92%).
Olivier Wulveryck proposes replacing local, isolated AI agents with a centralized 'agentic mesh' where agents charge one another via internal credits before executing any task. The key: combining the A2A protocol for orchestration and AP2 for payments, building an internal economy with cryptographic mandates.
A Hacker News thread argues that the greatest risk of AI is not a runaway superintelligence, but that governments and Big Tech capture it for the exclusive use of elites. The trigger: the recently announced 'regulation' of OpenAI's frontier models.
Brazilian digital transformation firm CI&T, listed on the NYSE under the ticker CINT, has become part of Anthropic's official Claude partner program. The move expands the ecosystem of certified integrators for Anthropic's AI model.
Dave Vellante and George Gilbert of SiliconANGLE argue that the AI industry is chasing the wrong goal: the race toward superintelligence ignores the real prize, enterprise AGI, that is, a single intelligence unique to each company built on its data, processes and tacit knowledge.
Robert Bateman publishes the Thought Tree framework on GitHub as an open 'handoff': an XML schema (TTML) to describe LLM flows as graphs of artifacts and operations. The project, still unfinished, hopes others will carry it on after the author's burnout.
The newsletter mentioned: The rise of the human–AI workforce