What public companies just told the SEC about AI, read straight from the filings. Every quote links the original document.
Data443 Launches Inline Enforcement Layer for AI Agents
Data443 Risk Mitigation (ATDS) filed an 8-K on October 14 announcing general availability of Vaikora, described as an AI runtime enforcement and control platform. Vaikora sits inline between an AI system and the actions it attempts, evaluating each tool call, function call, and data operation against policy at the moment of execution rather than logging violations afterward. Data443 describes two enforcement surfaces: agent actions, meaning tool and function calls the AI attempts to execute on customer systems, and data the AI attempts to read, write, or transmit.
Rather than inspecting AI behavior after the fact, Vaikora evaluates each action against policy at runtime and allows it, blocks it, or constrains it — and records what happened.
Data443 describes itself as a data security and privacy software company. Vaikora is its first product framed specifically around governing AI agent behavior at the action layer.
Agentic AI Integration Now Cited in CEO Succession Filings
ManpowerGroup (MAN) filed an 8-K on October 15 announcing the appointment of Becky Frankiewicz as incoming chief executive. The board's public statement named her leadership of "the integration of AI and agentic AI across the organization" as the central professional credential in the succession announcement. ManpowerGroup is a global workforce solutions and staffing company.
Frankiewicz has led ManpowerGroup's commercial strategy, driving the integration of AI and agentic AI across the organization and positioning the company and its global brands – Manpower, Experis, and Talent Solutions – for the next era of growth, innovation and transformation.
Separately, SciSparc (SPRC), a Nasdaq-listed company, filed a 6-K on October 15 announcing a "transformation into agentic AI-powered technologies, including cybersecurity and finance technologies." The available excerpt does not include financial terms or implementation details.
Defense and Drug Discovery Reach for Agentic AI in New Partnerships
Ondas (ONDS) disclosed in an October 13 8-K the formation of Regna AI, a joint venture with Palantir built on Ondas's SkyWeaver surveillance platform. The filing describes "embedded agentic AI with edge processing" as the technical foundation, with mission-specialized models targeting intelligence, surveillance, and reconnaissance in environments with limited connectivity. Ondas will be Regna AI's initial customer.
By combining embedded agentic AI with edge processing, SkyWeaver provides an operational foundation for Regna AI's mission-specialized models in multi-domain and degraded-connectivity environments.
Recursion Pharmaceuticals (RXRX) included forward-looking language in an October 15 8-K citing "anticipated benefits of our agentic AI tools, including expected improvements in scientific productivity, design and analysis cycle times, and clinical trial enrollment rates." The filing does not quantify those expected improvements.
Annual Filings Add AI-Enabled Cyberattacks and LLM Copyright as Named Risks
Marsh & McLennan included risk language in its Q3 2026 earnings filing that lists cyberattacks "enabled or enhanced by artificial intelligence" as a distinct threat category alongside conventional ransomware and supply chain compromises.
the increasing prevalence of ransomware, supply chain and other forms of cyber attacks, including those enabled or enhanced by artificial intelligence, and their potential to disrupt our operations, or the operations of our third party vendors, and result in the disclosure of confidential client or company information
Comtech Telecommunications (CMTL), in its fiscal 2026 10-K filed October 14, added a risk factor covering "infringement claims in connection with our use of artificial intelligence and machine learning." The language identifies LLM output specifically as the exposure source, warning that claims could arise from "third-party content contained in generation results from large-language models" used in product development and testing. The framing is narrower than the general AI liability language that appeared in 10-K filings a year ago, pointing toward a specific and ongoing legal concern rather than a category placeholder.