Beyond the Platform: Securing AI where your work actually happens | AlphaSense (part 2 of 3)

The security perimeter used to be something you could draw on a whiteboard. Data lived in your platform. If you wanted to work with it, you came to the platform. Anything that left was by design. SaaS eroded that, and AI has now pushed it well past where SaaS started. Today, work happens wherever your team has presence, and that "user" isn't always a person. It could be a subagent, an MCP connector, or an add-in running inside Excel. Larry Sica, VP of Product Security at AlphaSense, returns for Part 2 of the "From Assistant to Agent" security series, joined by Michelle Turner, Director of Product Marketing, for a technical working session on what changes when AI reaches out through connectors, add-ins, and APIs into the tools your team already uses. What you'll hear: - Why the perimeter has moved from device to network to presence, and what that means for how sessions cross device boundaries - The 6 attack vectors that just opened up or fundamentally changed: credentials, data movement, MCP trust boundaries, tool connectedness, supply chain amplification, and agent tenacity - Why zero trust is no longer optional, and how AlphaSense verifies the where, the who, and the why on every request - Why the 3 paths in (API, MCP connector, add-in) converge onto a single security channel, and how that avoids maintaining 3 separate control sets - 4 protections that hold true across every hop: end-to-end and at-rest encryption, entitlement inheritance from source systems, least-privilege short-lived tokens, and full audit logging - The 8 principles of what "good" looks like, grouped into what you assume about input, how you contain what runs, and how you know what happened - Why AlphaSense treats all uploaded content and retrieved data as data, never as instructions, to structurally resist prompt injection - How AlphaSense evaluates every third-party connector: identity model (per-user OAuth, not service accounts), data classification, and governance review - Why AlphaSense has denied connectors outright when they only supported service-account authentication - How AlphaSense in Excel and PowerPoint runs on managed endpoints inside your estate: deployment via Microsoft 365 admin center, no local admin required, no listening ports, and consent-based logging - Why zero data retention means the same thing everywhere at AlphaSense, and why Claude Code's ZDR is different from Claude Chat's ZDR by default 00:00 Welcome and Housekeeping 02:14 Meet Larry Sica, VP of Product Security 02:59 Beyond the Platform: What Changes When Work Leaves the Perimeter 03:34 From 4 Walls to Presence: The New Boundary 05:17 The 6 Attack Vectors That Just Changed 06:20 Supply Chain, Amplification, and Agent Tenacity 07:57 Why Zero Trust Is No Longer Optional 08:18 The 3 Questions: Where, Who, and Why 10:23 3 Doors, One Path: API, MCP, and Add-Ins 12:14 The 4 Protections Every Hop Must Carry 13:31 8 Principles: What We Assume, How We Contain, How We Know 13:51 Treating Every Input as Data, Never as Instructions 14:21 Isolating Execution, Ephemeral Sandboxes, and Least Privilege 15:11 Data Controls and Limiting Outbound Access 15:36 Why Zero Data Retention Has to Mean Zero 16:05 Observability and Layered Guardrails 17:32 How We Evaluate Every Connector 18:59 Building Our Own Connectors: Default Read-Only, Sandboxed Testing 20:01 Vendor Risk Reviews Before Any Code Is Written 20:26 The Distribution Problem: Marketplaces and Admin Approval 21:11 Excel and PowerPoint: Running on Managed Endpoints Inside Your Estate 22:19 Shared Machines, Session Isolation, and SSO 22:56 Network Access, Allow Lists, and Consent-Based Logging 23:28 What Happens When Things Fail 23:49 Certifications and Continuous Testing 24:07 Questions to Ask Any Vendor Using AI 25:29 Q&A: How AlphaSense's Security Posture Is Different 26:26 Q&A: What ZDR Actually Means Across Frontier Models 27:16 Q&A: Prompt Injection as an Input and Output Problem 27:57 Q&A: AlphaSense Direct vs Through Claude 30:04 Closing and Session 3 Preview Plus Q&A on how AlphaSense's security posture differs from competitors, why AlphaSense treats prompt injection as both an input and an output problem, and when to use AlphaSense directly vs through a general-purpose tool like Claude. For CISOs, product security leaders, IT and procurement teams, and anyone responsible for the enterprise AI stack. --- AlphaSense is a market intelligence and search platform used by the world’s leading companies and financial institutions. Learn more at https://www.alpha-sense.com