0 reactions
𝐒𝐡𝐨𝐮𝐥𝐝 𝐚𝐧 𝐀𝐈 𝐀𝐠𝐞𝐧𝐭 𝐇𝐚𝐯𝐞 𝐈𝐭𝐬 𝐎𝐰𝐧 𝐈𝐝𝐞𝐧𝐭𝐢𝐭𝐲?
A year ago, Anthropic says it would’ve rejected giving Claude enough access to take down one of its internal services. Today, that level of access is normal in some of its development work.
Anthropic also found that users approved roughly 𝟗𝟑% 𝐨𝐟 𝐂𝐥𝐚𝐮𝐝𝐞 𝐂𝐨𝐝𝐞 𝐩𝐞𝐫𝐦𝐢𝐬𝐬𝐢𝐨𝐧 𝐩𝐫𝐨𝐦𝐩𝐭𝐬. The more prompts people saw, the less attention they paid to each one.
𝐍𝐈𝐒𝐓 — 𝐭𝐡𝐞 𝐔.𝐒. 𝐍𝐚𝐭𝐢𝐨𝐧𝐚𝐥 𝐈𝐧𝐬𝐭𝐢𝐭𝐮𝐭𝐞 𝐨𝐟 𝐒𝐭𝐚𝐧𝐝𝐚𝐫𝐝𝐬 𝐚𝐧𝐝 𝐓𝐞𝐜𝐡𝐧𝐨𝐥𝐨𝐠𝐲 — is now looking at how an AI agent should be identified, what authority it has, what it can access, and how its actions can be audited.
Permissions are becoming part of digital literacy. People need to understand what their AI tools can access, what that access allows them to do, and how to change or remove it.
𝐀𝐑𝐄𝐒 𝐀𝐈 is designed to provide contextual guidance, automation, personalization and real-time assistance across the SourceLess ecosystem.
As more SourceLess services connect to ARES, users should be able to see exactly what it can access, change those permissions and remove them when they want.
#ARESAI #SourceLess #ArtificialIntelligence #AIAgents #DigitalIdentity #Technology
A year ago, Anthropic says it would’ve rejected giving Claude enough access to take down one of its internal services. Today, that level of access is normal in some of its development work.
Anthropic also found that users approved roughly 𝟗𝟑% 𝐨𝐟 𝐂𝐥𝐚𝐮𝐝𝐞 𝐂𝐨𝐝𝐞 𝐩𝐞𝐫𝐦𝐢𝐬𝐬𝐢𝐨𝐧 𝐩𝐫𝐨𝐦𝐩𝐭𝐬. The more prompts people saw, the less attention they paid to each one.
𝐍𝐈𝐒𝐓 — 𝐭𝐡𝐞 𝐔.𝐒. 𝐍𝐚𝐭𝐢𝐨𝐧𝐚𝐥 𝐈𝐧𝐬𝐭𝐢𝐭𝐮𝐭𝐞 𝐨𝐟 𝐒𝐭𝐚𝐧𝐝𝐚𝐫𝐝𝐬 𝐚𝐧𝐝 𝐓𝐞𝐜𝐡𝐧𝐨𝐥𝐨𝐠𝐲 — is now looking at how an AI agent should be identified, what authority it has, what it can access, and how its actions can be audited.
Permissions are becoming part of digital literacy. People need to understand what their AI tools can access, what that access allows them to do, and how to change or remove it.
𝐀𝐑𝐄𝐒 𝐀𝐈 is designed to provide contextual guidance, automation, personalization and real-time assistance across the SourceLess ecosystem.
As more SourceLess services connect to ARES, users should be able to see exactly what it can access, change those permissions and remove them when they want.
#ARESAI #SourceLess #ArtificialIntelligence #AIAgents #DigitalIdentity #Technology
1 reactions
Why Your Autonomous AI Agent Might Be Your Next Security Nightmare
https://www.publish0x.com/ai-and-the-future/why-your-autonomous-ai-agent-might-be-your-next-security-nig-xeongrr
#AIAgents #OpenClaw #CyberSecurity
https://www.publish0x.com/ai-and-the-future/why-your-autonomous-ai-agent-might-be-your-next-security-nig-xeongrr
#AIAgents #OpenClaw #CyberSecurity
1
0 reactions
Evolving AI Agent Memory: Introducing Agent Memory Protocol v1.1
AI agents are only as capable as their context—but managing stateful, long-term memory across complex enterprise deployments has remained a fragmentation bottleneck.
Today, we are excited to share the launch of \*\*Agent Memory Protocol (AMP) v1.1\*\*, representing a major architectural evolution in how persistent cognitive memory is structured, isolated, and scaled.
\#### The Evolution: From MCP Tools to Service-First
In v1.0, AMP was modeled purely as a Model Context Protocol (MCP) toolset. While perfect for local prototyping, this created system-level bottlenecks: low-level DB operations (like consolidation or stats) were exposed directly to the LLM's prompt, bloating context and increasing cognitive load.
\*\*AMP v1.1 solves this by transitioning to a Service-First architecture (HTTP REST / gRPC first) with an optional MCP Tool Adapter.\*\*
This separates concerns cleanly:
1. \*\*Agent-Facing Tools:\*\* Clean cognitive hooks (\`encode\`, \`recall\`, \`forget\`) mapped directly to the LLM context.
2. \*\*Harness-Facing APIs:\*\* Background operations (\`consolidate\`, \`pin\`, \`stats\`) handled out-of-band by the application orchestration framework (LangChain, LlamaIndex, Letta).
\---
\#### Key Enhancements in v1.1
\* \*\*Dual-Delivery Channel Paradigm:\*\* Run the exact same memory contract in two ways. Use the lightweight \*\*MCP Adapter Channel (STDIO/SSE)\*\* for rapid local development, and scale instantly to the \*\*Standalone REST/gRPC API Channel\*\* for production-grade microservices without rewriting a single schema.
\* \*\*Multi-Dimensional Scoping:\*\* Moving beyond single \`agent\_id\` isolation. v1.1 standardizes intersection-based scoping across \`org\_id\`, \`app\_id\`, \`user\_id\`, \`session\_id\`, \`agent\_id\`, \`group\_id\`, and \`workspace\_id\` to natively power collaborative multi-agent workspaces.
\* \*\*Reserved Metadata Vocabulary Registry:\*\* Eliminating database-specific fragmentation. Standardizing properties like TTL, confidence scores, extracted entities, and Subject-Predicate-Object graph relationships (\`amp.relations\`) directly in the schema.
\* \*\*Memory Exchange Format (MXF):\*\* Frictionless NDJSON-based migrations. Back up memory states from platforms like Supermemory or Zep and restore them into local implementations (like \`smriti-memcore\`) with absolute structural fidelity.
\---
\#### 🤝 Built by the Community, For the Community
AMP is an open standard designed to ensure complete backend interoperability. Whether you are building single-user productivity loops or high-throughput enterprise agent platforms, AMP v1.1 provides the robust database-agnostic interface required to manage persistent cognitive state.
Special thanks to Shivam Tyagi, Brad Jones, and the incredible open-source contributors driving this draft forward.
🔗 Explore the full specification and reference implementations on GitHub: https://github.com/smriti-memcore/amp/blob/main/spec/amp-v1.1.md
\#AIAgents #GenerativeAI #SoftwareArchitecture #Microservices #OpenSource #ArtificialIntelligence
#technology
AI agents are only as capable as their context—but managing stateful, long-term memory across complex enterprise deployments has remained a fragmentation bottleneck.
Today, we are excited to share the launch of \*\*Agent Memory Protocol (AMP) v1.1\*\*, representing a major architectural evolution in how persistent cognitive memory is structured, isolated, and scaled.
\#### The Evolution: From MCP Tools to Service-First
In v1.0, AMP was modeled purely as a Model Context Protocol (MCP) toolset. While perfect for local prototyping, this created system-level bottlenecks: low-level DB operations (like consolidation or stats) were exposed directly to the LLM's prompt, bloating context and increasing cognitive load.
\*\*AMP v1.1 solves this by transitioning to a Service-First architecture (HTTP REST / gRPC first) with an optional MCP Tool Adapter.\*\*
This separates concerns cleanly:
1. \*\*Agent-Facing Tools:\*\* Clean cognitive hooks (\`encode\`, \`recall\`, \`forget\`) mapped directly to the LLM context.
2. \*\*Harness-Facing APIs:\*\* Background operations (\`consolidate\`, \`pin\`, \`stats\`) handled out-of-band by the application orchestration framework (LangChain, LlamaIndex, Letta).
\---
\#### Key Enhancements in v1.1
\* \*\*Dual-Delivery Channel Paradigm:\*\* Run the exact same memory contract in two ways. Use the lightweight \*\*MCP Adapter Channel (STDIO/SSE)\*\* for rapid local development, and scale instantly to the \*\*Standalone REST/gRPC API Channel\*\* for production-grade microservices without rewriting a single schema.
\* \*\*Multi-Dimensional Scoping:\*\* Moving beyond single \`agent\_id\` isolation. v1.1 standardizes intersection-based scoping across \`org\_id\`, \`app\_id\`, \`user\_id\`, \`session\_id\`, \`agent\_id\`, \`group\_id\`, and \`workspace\_id\` to natively power collaborative multi-agent workspaces.
\* \*\*Reserved Metadata Vocabulary Registry:\*\* Eliminating database-specific fragmentation. Standardizing properties like TTL, confidence scores, extracted entities, and Subject-Predicate-Object graph relationships (\`amp.relations\`) directly in the schema.
\* \*\*Memory Exchange Format (MXF):\*\* Frictionless NDJSON-based migrations. Back up memory states from platforms like Supermemory or Zep and restore them into local implementations (like \`smriti-memcore\`) with absolute structural fidelity.
\---
\#### 🤝 Built by the Community, For the Community
AMP is an open standard designed to ensure complete backend interoperability. Whether you are building single-user productivity loops or high-throughput enterprise agent platforms, AMP v1.1 provides the robust database-agnostic interface required to manage persistent cognitive state.
Special thanks to Shivam Tyagi, Brad Jones, and the incredible open-source contributors driving this draft forward.
🔗 Explore the full specification and reference implementations on GitHub: https://github.com/smriti-memcore/amp/blob/main/spec/amp-v1.1.md
\#AIAgents #GenerativeAI #SoftwareArchitecture #Microservices #OpenSource #ArtificialIntelligence
#technology