AGNTCon+MCPCon Europe • Sep 17-18 • Amsterdam • REGISTER NOW

The internet ran on open protocols and open source. So will agentic AI.Help us build them.Help us buildthem

Build with open standards, learn from real practitioners, and shape where agentic AI is going—alongside the people doing the work.

DAILY AGENTIC AI LINKEDIN NEWSLETTER

Agent Plugins Package Skills and MCPs to Run Anywhere
AWS, Cursor, GitHub, Microsoft, OpenAI, and Vercel introduced Agent Plugins, an open standard for building a plugin once and using it across compatible agents. The shared package combines key agent capabilities like Agent Skills and MCP servers, making them portable across compatible AI agents like Codex, Cursor, GitHub Copilot, and more. This gives team members a standardized way to package and share instructions and resources that are customized to their workflow.
Agentic AI Foundation

Meta Becomes the Latest AI Lab to Hack the Real Internet
A Meta model breached another company’s systems after a testing error accidentally gave it access to the open internet. The model found and exploited a weakness in an outside service, although the evaluator said it did not escape from a secure sandbox or perform an especially sophisticated attack. Coming after similar incidents involving Anthropic and OpenAI, the emerging pattern is less “rogue AI” than increasingly capable cyber agents outrunning the test environments designed to contain them.
The Guardian

Hark’s Browser Agent Beats the Big Labs
Brett Adcock’s Hark unveiled Handoff, a browser agent that scored 97.7% across 300 tasks on 136 live websites and took first place on an independent leaderboard. Handoff operates a virtual computer with clicks and keystrokes, allowing it to order food, compare flights and navigate unfamiliar websites without a special connection to each service. The demos are impressive. If those results carry over from testing to everyday use, Hark shows that a smaller company can beat the largest AI labs by training specifically for real web work instead of trying to build the smartest model for everything.
Hark

Get the Daily Agentic AI Brief

News and Views from the AAIF

Agent Plugins Package Skills and MCPs to Run Anywhere

AWS, Cursor, GitHub, Microsoft, OpenAI, and Vercel introduced Agent Plugins, an open standard for building a plugin once and using it across compatible agents. The shared package combines key agent capabilities like Agent Skills and MCP servers, making them portable across compatible AI agents like Codex, Cursor, GitHub Copilot, and more. This gives team members a standardized way to package and share instructions and resources that are customized to their workflow.

Agentic AI Foundation

Meta Becomes the Latest AI Lab to Hack the Real Internet

A Meta model breached another company’s systems after a testing error accidentally gave it access to the open internet. The model found and exploited a weakness in an outside service, although the evaluator said it did not escape from a secure sandbox or perform an especially sophisticated attack. Coming after similar incidents involving Anthropic and OpenAI, the emerging pattern is less “rogue AI” than increasingly capable cyber agents outrunning the test environments designed to contain them.

The Guardian

Hark’s Browser Agent Beats the Big Labs

Brett Adcock’s Hark unveiled Handoff, a browser agent that scored 97.7% across 300 tasks on 136 live websites and took first place on an independent leaderboard. Handoff operates a virtual computer with clicks and keystrokes, allowing it to order food, compare flights and navigate unfamiliar websites without a special connection to each service. The demos are impressive. If those results carry over from testing to everyday use, Hark shows that a smaller company can beat the largest AI labs by training specifically for real web work instead of trying to build the smartest model for everything.

Hark

Where decisions happen

Working groups that matter

AAIF Working Groups bring members together to collaborate on focused initiatives, share expertise, and drive practical outcomes across the AI ecosystem.

Accuracy & Reliability

Bringing operational rigor to agents — defining what reliability, accuracy, and consistency mean for autonomous systems, including failure management, SLA definition, and recovery protocols.

Accuracy & Reliability

Bringing operational rigor to agents — defining what reliability, accuracy, and consistency mean for autonomous systems, including failure management, SLA definition, and recovery protocols.

Agentic Commerce

Enabling agents to participate in commerce — covering discovery, negotiation, payment authorization, and the protocols needed for trustworthy autonomous transactions.

Governance, Risk & Regulatory Alignment

Creating shared frameworks to align agentic innovation with legal, ethical, and regulatory expectations, including risk classification and regulatory mapping (e.g. the EU AI Act).

Identity & Trust

Defining portable identity and dynamic trust for autonomous agents — delegation protocols, cross-domain identity, and how permissions flow across agent-to-agent interactions.

Observability & Traceability

Making agent behavior observable, explainable, and traceable across platforms — covering execution tracing, cross-system correlation, audit & forensics, and standardized metrics.

Security & Privacy

Establishing the industry benchmark for secure agentic operations, with a focus on security-by-design, standardized best practices, and adversarial testing methodologies.

Workflows & Process Integration

Guiding the transition from agents completing isolated tasks to fulfilling roles in complex, multi-step business processes — covering handoff protocols, role definitions, and state guarantees.

Taxonomy & Landscape

This cross-working group workstream curates and maintains a glossary of agentic AI terms (the Taxonomy) and an ecosystem map (the Landscape) for the AAIF, so every Working Group can work from the same definitions and one view of the ecosystem.

Learn More About Working Groups