AI agents

Executives and engineers at two Silicon Valley AI events said building and running AI agents remains difficult and costly. Kevin McGrath, CEO of Meibel, warned that routing every task through a large language model wastes tokens, while Google engineer Deep Shah focused on managing the inference costs of large agent fleets. ThinkingAI co-founder Chris Han said OpenClaw is too complicated and too prone to security flaws for enterprise use.

OpenAI released a major update to Codex, its coding agent used by more than 3 million developers weekly. The update adds background computer use, where multiple agents can operate Mac apps in parallel by seeing, clicking and typing with their own cursor, plus an in-app browser, image generation with gpt-image-1.5, memory of user preferences, and more than 90 plugins that combine skills, app integrations and MCP servers.

Pentera's AI Security and Exposure Report 2026 says every surveyed CISO reports AI already in use across their organizations. The article argues that security validation needs AI for adaptive testing, but that fully agentic systems, where AI reasoning governs execution end to end, undermine the repeatability that structured security programs require. It proposes a hybrid model in which deterministic logic defines attack chain execution and AI adapts payloads within that structure.

FinBot is a hands-on agentic AI capture-the-flag platform from the OWASP GenAI Security Project's Agentic Security Initiative, described as the "Juice Shop for Agentic AI." It simulates a multi-agent vendor management platform with LLM-driven onboarding, fraud detection, invoice processing and communications, and its challenges map to the OWASP Top 10 for LLM Applications, the OWASP Top 10 for Agentic Applications, CWE and MITRE ATLAS. The source positions it as a companion to the Agentic Top 10 framework rather than a replacement.

Swedish vendor Curity announced Access Intelligence, an extension to its Identity Server API IAM platform, to secure autonomous AI agents. Its approach treats agents as a special type of application, issuing OAuth tokens that carry the agent's purpose and intent, with access granted per action at runtime rather than through static permissions.

This AWS blog post explains how to secure AI agents and coding assistants that access AWS resources through the Model Context Protocol (MCP). It argues that agents can do anything their granted entitlements allow, so IAM permissions must be designed as deterministic controls, and it presents three IAM security principles with policy examples.

Cloudflare is making OpenAI frontier models, including GPT-5.4, available to millions of customers within Agent Cloud, a platform for deploying AI agents that handle tasks such as responding to customers, updating systems and generating reports. The Codex harness is now generally available in Cloudflare Sandboxes and will come to Workers AI in the near future.

Microsoft describes the agentic SOC, a security operations model that moves from reacting to incidents toward anticipating attacker movement and reshaping the environment to cut off their paths. The model pairs a threat protection platform with built-in autonomous defense and AI agents that assist analysts with investigation and prioritization. The article is the opening of a series and points to a new whitepaper, The agentic SOC: Your teammate for tomorrow, today.

Microsoft has released the Agent Governance Toolkit, an open-source project that monitors and controls AI agents during execution. It adds a runtime security layer that enforces policies against issues such as prompt injection and maps to OWASP's top 10 risks for agentic systems. The toolkit is in public preview under an MIT license, with components in Python, TypeScript, Rust, Go and .NET.

The source text is a bibliographic citation for an article titled "SBOMs into Agentic AIBOMs: Schema Extensions, Agentic Orchestration and Reproducibility Evaluation," published in Digital Threats: Research and Practice, Volume 7, Issue 2, pages 1-35, June 2026. It contains no abstract, method, or findings, so the research question and results cannot be stated from this text.

The source argues that the cybersecurity response to AI-enabled nation-state threats cannot be incremental and must instead be architectural. It is a short opinion piece with no further details on specific attacks, products or figures.