WitnessAI

No entries match these filters. Clear filters

1. MCP authentication and authorization: OAuth for AI agents (2026-09-28) Sep 28, 2026 · issue 088

   WitnessAI adds MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. server/tool discovery, approved-tool enforcement, and bidirectional content inspection for enterprise AI agent governance.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

   • Adds an organization-wide approved-tool list for MCP servers and tools, enforced at the network level before agent calls execute across IDE, chat, and agent surfaces via Agentic Control.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠
   • Adds MCP server and tool visibility through the Observe module, identifying agentic sessions by inspecting tool advertisements in traffic payloads across Claude Desktop, VS Code with AI extensions, and local agents built on LangChain, LlamaIndex, CrewAI, and custom code.
   • Adds bidirectional content inspection via the Protect module's runtime APIs, evaluating agent inputs before processing and responses before delivery, with malicious prompt detection validated in production across more than 100 LLMsLLMsA large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code..
   • Adds unified audit trails that correlate agent actions and blocked tool calls to a triggering human identity, covering both agent and employee activity routed through the platform.
   • Supports intent-based machine learning classification of conversational context to drive Control module enforcement decisions: allow, warn, block, or route to an approved internal model.

   5 details in Everything

2. AI data classification: how it works (2026-09-13) Sep 13, 2026 · issue 076

   WitnessAI adds intent-based AI data classification with agentic controls, MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. allow-lists, and data tokenization across the full prompt lifecycle.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

   • Adds Observe module for network-level discovery of AI tools and MCP connections — including native apps like Windows Copilot and Microsoft 365 — cataloging thousands of AI applications without endpoint clients or browser extensions, and attributing captured agent actions to a human identity.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠
   • Adds Agentic Control to enforce approved-tool and MCP allow-lists, block unauthorized tool calls before execution, perform pre-execution checks on approved tool calls, apply runtime controls to agent activity, and maintain a full audit trail of governed agent and tool-call activity across more than 100 LLMLLMA large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. types.
   • Adds Protect runtime guardrails to inspect prompts and filter harmful or off-policy responses in AI interactions routed through the platform, covering both inputs and generated outputs for PIIPIIPersonally Identifiable Information — data that can identify a specific individual, such as names, email addresses, or SSNs. Cyber tools flag or redact it to limit exposure and meet privacy regulations. and sensitive data per NIST Generative AI Profile guidance.
   • Adds Control capabilities to enforce four-tier intelligent policies — allow, warn, block, and route — on AI traffic routed through the platform, enabling proportionate enforcement over prompt, upload, response, and tool-call interactions.
   • Adds data tokenization that replaces classified sensitive values (such as SSNs) before prompts leave the enterprise and policy-enforced detokenization that restores original values for authorized users in model responses.

   +3 more show less

   • Adds intent-based machine learning classification that evaluates conversational context and purpose to detect sensitive data, suspicious behavior, and policy violations that keyword and regex matching cannot represent, with confidence scores adjusted by contextual signals.
   • Adds routing action within Control to redirect sensitive queries (such as non-public drug research or financial data) from external models to approved internal models based on classified intent.
   • Produces auditable records of what employees and agents sent to which model and what came back, providing evidence for AI risk reviews under frameworks such as DORA and the EU AI ActEU AI ActEU regulation that classifies AI systems by risk level and imposes compliance obligations—documentation, transparency, human oversight—on developers and deployers, shaping how AI-based cyber tools are built and sold in Europe..

   8 details in Everything