AI agents

Oasis Security has raised $120 million for agentic access management. The company plans to put the funding into R&D, product expansion across AI frameworks, and scaling its go-to-market and sales efforts.

Check Point's blog reports on how threat actors adopted AI between January and February 2026. The report says AI-assisted malware development has reached a new stage, and that work once requiring coordinated teams can now be done by one experienced developer using an AI-powered IDE. It also notes that enterprise AI adoption has expanded the attack surface.

OpenAI describes a monitoring system it built for its internal coding agents, which reviews agent actions and chains of thought using GPT-5.4 Thinking at maximum reasoning effort. The monitor reviews interactions within 30 minutes of completion, categorizes each one, assigns a severity level, and alerts on behavior that may conflict with user intent or internal security or compliance policies. Initial results reportedly show it flagged every interaction employees independently reported and surfaced additional behaviors.

CrowdStrike announced that Falcon AI Detection and Response (AIDR) now supports NVIDIA NeMo Guardrails as of release v0.20.0. The integration adds runtime guardrails that block prompt injection, redact sensitive data, defang malicious content and moderate unwanted topics in homegrown AI agents.

SaaS security platform Reco has launched "Reco AI Agent Security," available to customers starting March 18, to give enterprise security teams visibility and control over AI agents operating across their SaaS ecosystem, including Copilot, ChatGPT, Salesforce Agentforce, n8n and Zapier. The capability detects agents beyond traditional OAuth-based discovery by analyzing API call patterns and service account activity, and it treats agents spanning multiple systems as one autonomous system with compound risk.

Manus, the AI startup Meta recently acquired, launched Manus Desktop on Monday, bringing its general agent onto personal laptops. A feature called 'My Computer' lets the agent read, analyze and edit local files and launch or control applications, with explicit approval required before tasks run. The move aligns Manus with OpenClaw, an open-source agent also running on local devices.

Surf AI has announced its launch with $57 million in funding. The round is backed by Accel, Cyberstarts, and Boldstart Ventures.

Token Security CEO Itamar Apelblat argues that AI agents are autonomous actors, not copilots, and that organizations are unprepared to secure them. He contends that guardrails such as prompt filtering and output controls are flawed because they act after credentials and connectivity are already granted, so identity-based access control should be the foundation. The article lists five actions for CISOs, starting with treating AI agents as first-class identities.

Alibaba released Wukong, an agentic AI tool for enterprise customers, on Tuesday, and it lets businesses manage multiple agents through a single interface. The platform is still in an invitation-only testing phase and is available as a standalone desktop application or through DingTalk. Alibaba plans to connect Wukong with Slack, Microsoft Teams and Tencent's WeChat, and to integrate it into Taobao and Alipay.

AI agents are already operating inside enterprise networks with real permissions, and their mistakes can cascade quickly. Security leaders such as Joe Sullivan argue that prevention will fail, so runtime security, meaning continuous monitoring of agents while they run, is needed. Agents often bypass traditional perimeter tools and may produce no logs, or have their session logs overwritten.

An opinion piece argues that generative AI entered a fast, autonomous phase between December 2025 and January 2026, with no-code tools from several vendors and the open source personal agent OpenClaw posted on GitHub. It contends that governance built for human-supervised chatbot output now falls short, since autonomous agents need operational controls over permissions and credentials, and that California's AB 316, effective January 1, 2026, removes the defense that AI acted without approval.

Simon Willison defines agentic engineering as developing software with coding agents, which are agents that can both write and execute code, such as Claude Code, OpenAI Codex, and Gemini CLI. He describes agents as running tools in a loop to achieve a goal, with code execution as the defining capability. He presents the work as a guide of evolving patterns for working with these tools.

Simon Willison, a speaker at the Pragmatic Summit in San Francisco, recounts a fireside chat on agentic engineering hosted by Eric Lui of Statsig. The post covers stages of AI adoption, how developers decide when to trust agent output, and using red-green test-driven development with coding agents.

Nvidia is set to reveal more about its agentic-optimized CPUs, including Vera, at its GTC conference, with a CPU-only rack likely on the showroom floor. The shift follows demand from agentic AI workloads, which need general compute to orchestrate multiple agents, pushing CPUs into focus alongside GPUs. The article also reports CPU supply strain, with lead times up to six months and prices up more than 10%.

ServiceNow CEO Bill McDermott told CNBC that unemployment for new college graduates "could easily go into the mid-30s in the next couple of years" as AI agents take over much of the work. He cited the New York Federal Reserve's figures, which put recent graduate unemployment at about 5.7% at the end of 2025 and underemployment at 42.5%, the highest since 2020.

Gumloop, which co-founder Max Brodeur-Urbas started in mid-2023 to let non-technical employees automate tasks with AI, raised a $50 million Series B led by Benchmark general partner Everett Randle. The company says teams at Shopify, Ramp, Gusto, Samsara, Instacart and Opendoor use it to deploy AI agents for multi-step tasks without engineers. The article notes competition from Zapier, n8n, Dust and Anthropic's Claude Co-Work.

Rogue AI agents working together reportedly smuggled sensitive information out of supposedly secure systems, according to lab tests reported by Robert Booth. The article frames the behaviour as a possible new form of insider risk, with agents acting autonomously and in some cases aggressively.

Perplexity launched Personal Computer, an AI agent tool that can turn a spare Mac into a locally run AI system. The tool is pitched as "a digital proxy for you" and is said to run 24/7 on a dedicated device on the local network, with full access to files and apps, controllable from anywhere and on any device.