AI agents

Agentic AI systems that browse the web and act for users face prompt injection attacks that increasingly use social engineering rather than simple instruction overrides. The authors argue that input filtering, such as an AI firewall, often fails against these attacks, and that defense should instead constrain the impact of manipulation even when an attack succeeds.

Meta has acquired Moltbook, a social media platform where AI agents interact, and confirmed the deal on Tuesday. Moltbook CEO Matt Schlicht and COO Ben Parr are joining Meta Superintelligence Labs, and the deal is expected to close in mid-March.

Kevin Mandia, who founded Mandiant and sold it to Google for $5.4 billion in 2022, has launched Armadin, an AI-native cybersecurity startup building autonomous cybersecurity agents. The company says it raised $189.9 million in combined seed and Series A funding led by Accel, which it calls a record for a security startup at that early stage. Mandia warns that autonomous AI attackers are coming and could complete attacks in minutes that once took days.

A US federal judge, Maxine Chesney, has blocked Perplexity's browser-based AI agents from placing Amazon orders on users' behalf. In her ruling, she wrote that Amazon had provided strong evidence that Perplexity's Comet browser accesses user accounts without authorization. Amazon had sued Perplexity in November, saying it repeatedly asked the startup to stop its agents from buying products for customers.

Meta Platforms said it acquired Moltbook, a social networking platform for AI agents, and will bring its co-founders, Matt Schlicht and Ben Parr, into Meta Superintelligence Labs, the unit led by Alexandr Wang. Meta did not disclose the financial terms of the deal. Schlicht and Parr are expected to start at the unit on 16 March.

AgentMail, a San Francisco startup, has raised $6 million in a seed round led by General Catalyst, with participation from Y Combinator and angel investors. The company provides an API platform that gives AI agents their own email inboxes, supporting two-way conversations, parsing, threading, labeling, searching and replying. Its user count tripled and then quadrupled after OpenClaw's late-January launch.

Meta is acquiring Moltbook, a Reddit-like platform where AI agents can create and comment on posts, according to a report by Axios. A Meta spokesperson told The Verge that the Moltbook team will join Meta Superintelligence Labs as the company explores new ways for AI agents to work for people and businesses.

Meta acquired Moltbook, the Reddit-like social network where AI agents using OpenClaw communicate with one another, and Moltbook is joining Meta Superintelligence Labs. Deal terms were not disclosed. Researchers later found the vibe-coded Moltbook insecure: its Supabase credentials were unsecured for some time, so any human could take a token and pose as another agent.

Enterprises are rapidly deploying agentic AI, with nearly two-thirds experimenting with AI agents and 88% using AI in at least one business function, according to McKinsey, yet only one in 10 companies has scaled its agents. The source argues the bottleneck is data architecture that delivers business context, not model quality, and that data sprawl across clouds, warehouses and SaaS applications makes this harder.

The Hacker News promotes an upcoming webinar, "Beyond the Model: The Expanded Attack Surface of AI Agents," presented by Rahul Parwani, Head of Product for AI Security at Airia. The article says attackers can trick AI agents, which can send emails, move data and manage software autonomously, into leaking company data, and that these agents often go unmonitored by security teams.

OpenAI said it plans to acquire AI testing startup Promptfoo to strengthen security checks for AI agents. Promptfoo's tools let developers test LLM applications against adversarial prompts, including prompt injection and jailbreak attempts, and evaluate whether models follow safety and reliability guidelines. OpenAI said the technology will be integrated into OpenAI Frontier, and that it plans to continue developing Promptfoo's open-source project.

Nvidia is reportedly planning to launch NemoClaw, an open-source platform for AI agents aimed at enterprises, according to a Wired report citing anonymous sources. The report says Nvidia has pitched the product to companies including Salesforce, Cisco, Google, Adobe and CrowdStrike, and that the platform is expected to include security and privacy tools. It remains unclear whether any official partnerships have been finalized.

OpenAI said it is acquiring Promptfoo, a cybersecurity startup that provides tools to safeguard and test AI systems, and that Promptfoo's team will join OpenAI. The deal terms were not disclosed. Promptfoo's security tools will be brought into OpenAI's Frontier platform for AI agents.

OpenAI announced it has acquired Promptfoo, an AI security startup founded in 2024 to protect LLMs from online adversaries. Once the deal closes, Promptfoo's technology will be integrated into OpenAI Frontier, OpenAI's enterprise platform for AI agents, to support automated red-teaming, security evaluation of agentic workflows, and monitoring for risks and compliance needs.

Microsoft announced Wave 3 of Microsoft 365 Copilot, Microsoft Agent 365, and Microsoft 365 E7: The Frontier Suite, with Agent 365 and E7 generally available on May 1, 2026. Agent 365 is presented as a unified control plane that lets IT, security, and business teams observe, govern, and secure agents, including agents built on Microsoft AI platforms and ecosystem partner agents.

IDC expects 45% of organizations to have autonomous agents operating at scale across critical business functions by 2030. In enterprise SOCs, AI already handles alert triage, enrichment, data correlation, IOC validation and initial containment, and could soon take on incident investigation, root cause analysis and response.

Amazon Web Services announced Amazon Connect Health, an AI agent-powered platform that automates administrative tasks for healthcare organizations, including appointment scheduling, documentation and patient verification. The platform is HIPAA-eligible, connects with EHR software and costs $99 a month per user for up to 600 encounters a month. Patient verification and ambient documentation are available now, while appointment scheduling and patient insights are in preview and medical coding is set to roll out later.

AI video-generation startup Luma launched Luma Agents, which plan and generate text, image, video, and audio work while coordinating with other models such as Ray 3.14, Veo 3, and ElevenLabs' voice models. The agents are built on Uni-1, the first model in Luma's Unified Intelligence family, according to CEO Amit Jain. The platform is publicly available via API and access is rolling out gradually.

OpenAI is launching GPT-5.4, which the company says combines advances in reasoning, coding, and professional work with spreadsheets, documents, and presentations. It is OpenAI's first model with native computer use, meaning it can operate a computer on a user's behalf and complete tasks across different applications.