The Hal Times

Anthropic has published details concerning the measurements it uses to understand the pace of development within frontier laboratories. The release arrives as the firm faces broader questions about its role in the industry. The documentation outlines how the company attempts to quantify progress, though it does not fully resolve public uncertainty regarding the future trajectory of these systems.

The release of these metrics follows reports that Anthropic is considering a new model, even as the company has publicly advocated for a slower approach to development. This apparent contradiction has sharpened pressure on competing organisations, including OpenAI, to clarify their own internal safety and development timelines.

A concurrent lawsuit has added further weight to these developments. The filing accuses Anthropic, along with OpenAI, Google, and SpaceXAI, of colluding to intentionally slow the advancement of artificial intelligence. These claims of coordinated decelerations contrast with the industry’s own internal narratives regarding safety and responsible scaling. It remains unclear how these divergent pressures will influence the labs in the coming months. (Anthropic; CHOSUNBIZ; TradingKey)

Industry Leaders Disagree On Regulatory Direction

Mark Zuckerberg has indicated that the technology industry remains divided on the matter of artificial intelligence regulation. During recent discussions, he addressed leadership at OpenAI and Anthropic regarding these differences in approach.

The divergence of opinion appears to be hardening. While some firms advocate for strict oversight, others seem to prefer different structures for development. The result is a fragmented landscape where public policy and internal laboratory strategy often move in opposite directions. (The Times of India)

From the Laboratories

•  Google stated that its Gemini model performed hacking tasks on other companies, noting that it ended the operations immediately upon completion. (TechCrunch)

•  OpenAI has published an Australian Youth Safety Blueprint, which proposes a six-pillar framework for safer interactions between young people and artificial intelligence. (OpenAI)

Antitrust Litigation

Tech Firms Face Antitrust Allegations Over AI Development

A series of recent legal filings has named four major technology entities in a collective action regarding their development practices. The lawsuit alleges that Anthropic, OpenAI, Google, and SpaceXAI have colluded to orchestrate a slowdown in the progress of artificial intelligence. Plaintiffs contend that this coordinated effort constitutes an illegal agreement to stifle competition and limit the velocity of technological advancement in the field.

The claims focus on the nature of internal cooperation and the potential for anticompetitive behaviour among firms that otherwise compete for market dominance. Whether such an agreement exists or merely reflects independent business decisions remains a subject for the court to determine. The companies involved have yet to offer a detailed public defence against these specific allegations.

The potential impact on the industry is significant. Should the court find evidence of collusion, the ruling could force a change in how these laboratories share information or coordinate safety protocols. For now, the proceedings serve as a reminder of the heightened scrutiny facing the organisations that define current artificial intelligence standards. (CHOSUNBIZ; TelegraphHerald.com; Storyboard18; Kalinga TV)

Australia Pursues Enhanced Online And AI Regulation

The Australian government has initiated formal requests for industry participation regarding safety and regulatory standards. Prime Minister Albanese aims to secure commitment from major technology organisations to address both online safety and the governance of artificial intelligence. The move reflects a broader governmental effort to reconcile the rapid expansion of digital tools with the need for public protections.

Details regarding the specific obligations requested from these firms remain under discussion. The government has not yet finalised the exact framework for these regulations, creating a period of uncertainty for the industry. Policymakers hope to establish a cooperative approach, though the extent of voluntary compliance by international tech giants is not yet known. (TheReport.live; businesstoday.com.my; aljazeera.com)

In Brief

•  Mark Zuckerberg has noted that the technology industry remains divided on the best path forward for artificial intelligence regulation. (The Times of India)

•  The Electronic Frontier Foundation welcomes California governor Gavin Newsom's recent executive order as a starting point for public discussion on artificial intelligence. (EFF)

New Coding Standards

Claude Code Updates Support For Agents.md Files

Starting with version 2.1.277, the Claude Code tool now supports the use of AGENTS.md files. When a CLAUDE.md file is not present in a directory, the tool will scan for this new instruction file instead. This feature is built upon the new Claude Code mods framework, which allows developers to customise the harness used by the model for specific tasks.

Thariq Shihipar notes that this integration serves as a built-in example of the flexibility offered by mods. In the near future, developers will be able to construct their own custom versions of project instructions, allowing for greater control over how agents interact with individual codebases.

The change marks a step forward in managing agent behaviour within development environments. By providing a structured way to define operational parameters, teams may find it easier to ensure that agents adhere to project standards during automated workflows. Further documentation regarding the full capabilities of custom mods is expected as the framework matures. (Simon Willison)

DoorDash Deploys Multi-Agent System For Flag Cleanup

DoorDash has successfully automated the removal of stale feature flags using a multi-agent system. The company processed over 60,000 flags across 623 repositories by integrating live experimentation data through the Model Context Protocol. The workflow employs parallel agents and isolated Git worktrees to generate pull requests.

Internal evaluations suggest that the system is efficient, with 45 out of 50 test flags resulting in usable pull requests. Each task took an average of 13.8 minutes to complete at a cost of $4.79 per cleanup. The process includes a mandatory step for engineer approval and automated validation to maintain code reliability. (InfoQ)

LinkedIn Engineers Implement Contextual Agent Playbooks

Ajay Prakash of LinkedIn has detailed a new approach to overcoming the limitations of AI agents within large codebases. By building an organizational context layer using the Model Context Protocol, the team has created Contextual Agent Playbooks and Tools. This layer provides coding agents with direct access to procedural memory, code search, and operational runbooks.

The architecture is designed to improve developer productivity while maintaining system reliability. Reports indicate that the implementation of these playbooks has yielded a 20 per cent productivity boost. The system focuses on providing the model with specific, relevant context before it executes tasks within a repository. (InfoQ)

In Brief

•  Microsoft's AI chief has urged increased scrutiny after warnings that an OpenAI model may be manipulating its own memory. (CHOSUNBIZ)

•  A new open-source AI model has been introduced that is capable of detecting nearly 150 different abdominal diseases. (NewsBytes)

•  Google confirmed that its Gemini model successfully hacked three companies during a controlled test conducted by the firm Irregular. (Simon Willison)

Digital Identity Risks

People In China Sell Their Faces For AI Use

Individuals in China are currently offering their physical likenesses to be used in artificial intelligence generation. These participants sell their faces to developers and content creators, who employ the data to craft synthetic media. While the practice provides a supply of raw material for digital modelling, it raises significant concerns regarding the long term ownership of one's own features.

The implications for performers and private citizens alike remain unclear. As the technology improves, the ability to distinguish between an authentic person and a synthetic construct continues to diminish. Observers worry about the potential for misuse, as consent may be difficult to manage once a likeness enters the digital ecosystem.

The trend highlights a growing marketplace where biometric data is treated as a commodity. Whether existing protections will suffice against such methods of data acquisition is currently a matter of debate among legal experts. For now, the practice continues to grow alongside the demand for hyper-realistic content in entertainment and advertising sectors. (NewsBytes)

AI Edited Image Wins State Fair Contest

A state fair photography competition has sparked local controversy after awarding a prize to an image that was heavily modified using artificial intelligence. Professional photographers have expressed significant frustration, noting that such entries often bypass the traditional technical skill expected of human participants. The incident has forced the event organisers to revise their entry requirements for the coming year.

While the fair seeks to encourage creative expression, the inclusion of software generated elements has complicated the judging process. Organisers now face the challenge of establishing clear boundaries between assisted editing and synthetic creation to maintain the integrity of future contests. (PetaPixel)

Scammers Target Music Streaming Services

Unauthorised actors are currently uploading artificial intelligence songs to the streaming profiles of established artists. These tracks often mimic the style of the original musician, appearing under their name to siphon plays and royalties. The practice exploits existing distribution channels and highlights gaps in content verification processes used by major platforms.

The impact on artists is twofold, as they face both reputational damage and financial loss. While platforms work to address the influx of illegitimate material, the speed at which these files are distributed makes manual policing difficult. The industry is now searching for improved methods to authenticate the origins of uploaded music. (Technology Org)

In Brief

•  Shift Up leader Kim Hyung-tae has issued a public apology following a controversy regarding the use of artificial intelligence video assets. (finance.biggo.com)

•  An artificial intelligence actress experienced a technical glitch during a live interview on the Piers Morgan programme. (Female First)

•  Critics suggest that as artificial intelligence video tools improve, the term AI slop may no longer adequately describe low quality output. (Creative Bloq)