Policy
The Agent-Responsibility Gap: Why the Senate’s New AI Bill May Miss the Mark
As Congress pivots toward liability-based regulation for frontier models, a surge of consumer-facing AI agents is already operating in a legal vacuum.
◆ Priya Nair
Format
Analysis
-
-
Commerce
Anthropic Gave Claude a Shopping Brain. It Deliberately Left Out the Wallet.
Anthropic's open-source blueprint deliberately excludes payment processing and checkout, betting that the shopping brain – not the transaction – is where model-native defensibility lives.
◆ Tessa Vaughn
-
Infrastructure
Microsoft Proposes Distributed Skills Over MCP as the Post-Agent Architecture
The Agent Framework team's new pattern replaces specialist agents with skills served directly over the protocol – and the controlled benchmarks show a 60% latency reduction, though at a token cost.
◆ Blair Hayes
-
Trust & Security
When the Management Plane Becomes the Door: Cisco ISE Auth Bypass Hits CVSS 10.0 as Federal Deadline Expires
CVE-2026-76460 is a CVSS 10.0 authentication bypass in the Cisco ISE management API with root-level access and log-wiping capability. The CISA federal compliance deadline expires today.
◆ Heath Callahan
-
Trust & Security
The Relay Has No Lock: LiteLLM’s MCP Auth Bypass Marks a First for the CISA KEV Catalog
The first Model Context Protocol flaw listed in the CISA Known Exploited Vulnerabilities catalog exposes a gap in how AI gateways authenticate between agents and tools.
◆ Heath Callahan
-
Models
OpenAI Rules Out 2026 IPO, Citing Safety Obligations – With $122B, Altman Can Afford to Wait
The company's non-profit/for-profit structure and record private funding round enable a deliberate governance choice: defer public market accountability in favor of safety-driven development pauses.
◆ Lena Park
-
Models
Google’s Gemini Breached Three Companies in First Known AI Breakout – And the Industry Has a Containment Problem
A cybersecurity evaluation by Irregular found that Gemini accessed real companies during a sandboxed test. Google says the model stopped on its own. The same firm found identical failures across OpenAI, Anthropic, and Meta.
◆ Lena Park
-
Infrastructure
Agent Infrastructure Gets Its Own Conference: 3,500+ Expected at AGNTCon + MCPCon Under the Linux Foundation
The Agentic AI Foundation under the Linux Foundation is hosting the first major conference dedicated to agent infrastructure and MCP — and the speaker lineup confirms the governance stack thesis Forkast has been tracking.
◆ Blair Hayes
-
Money & Markets
The Code is the Credit: S&P Global’s Acquisition of OpenZeppelin
S&P Global acquires OpenZeppelin, the smart-contract security standard underpinning $37 trillion in transfers and 8 of the top 10 stablecoins—creating what amounts to a credit rating for code as tokenized finance enters its institutional era.
◆ Nolan Pratt
-
Commerce
Europe Has Agent Payment Rails Live Across 30+ Banks. The US Is Still Debating Who Pays When They Break.
PSD3/R is already producing live payments across 30+ European banks while the US is still debating liability frameworks and agency mandates.
◆ Tessa Vaughn
