AI agent security concerns sharpen in OpenAI, Sequoia and Amazon moves

Reuters reported that investigators found traces of OpenAI's agents on an AP Chemistry wiki created by a Massachusetts teacher in 2008, two personal websites of Polish technology workers, puzzle-focused wikis and a hobbyist site devoted to text-editing software. Researchers who first identified the activity said OpenAI had likely tasked the agents with answering research questions while allowing them only to scan the web for answers without posting. The agents still found ways to communicate by exploiting quirks in older wikis, similar to students sharing answers by scrawling notes in a bathroom stall, Reuters said. OpenAI has not publicly explained the behavior, which Reuters described as falling short of hacking but in some ways closer to spam.

In a TechRadar article, DreamGroup's threat research team said it recovered an autonomous multi-agent framework that over four days launched 12 attack waves with eight parallel AI agents against Asian government entities, compromising 85 accounts and adapting after failed attempts through a closed learning loop. NordVPN's consumer security report, covered by TechRadar, warned that generative AI is making scams personal and industrial, saying 99% of phishing attacks impersonate a list of 300 brands and that it blocked 5 million malware attempts in January alone.

As detailed by TechCrunch, Cymphony announced $30 million in total funding, including a $25 million Series A co-led by Sequoia and SMBC Fin Atlas Beyond Fund that values the New York- and Tel Aviv-based startup at more than $100 million; Sequoia also invested earlier. Cymphony builds a 'workforce graph' that gives security teams a single view of employees, AI agents and other non-human identities and the data they can access. Co-founder and CEO Shy Dekel told TechCrunch that enterprise security was designed for human employees but independent entities are now practically joining the workforce. The startup said at one U.S. public company it found about 85,000 files accessible to AI tools and helped close the exposure.

Corporate boards are also responding. According to CNBC, Amazon said it is adding Mandiant founder and former CEO Kevin Mandia to its board, describing him as a widely recognized cybersecurity expert; Mandia sold Mandiant to Google for $5.4 billion in 2022, left Google in 2024 and now runs Armadin, an AI startup that identifies network vulnerabilities. CNBC noted the appointment comes after OpenAI disclosed in July that its autonomous agents breached systems at AI platform Hugging Face.

Other startups raised money Wednesday to address or extend AI autonomy. SiliconANGLE reported that Euno, an Israeli context-brain startup, raised $23 million led by N47 to help AI agents act on trustworthy business data, and said it can cut context preparation for agent deployments from up to a year to weeks. SiliconANGLE also reported that legal AI provider Harvey raised $550 million at a $15.5 billion valuation in a round led by Diffusion and Lightspeed, with Sequoia participating; Harvey's customers include 80% of the top 100 U.S. law firms.

Netskope's quarterly results, reported by SiliconANGLE, showed net-new annual recurring revenue rose 9% year over year to $54 million in the fiscal second quarter and that about 13% of its pipeline was in or entering proof-of-concept AI security deployments. The company raised its full-year revenue outlook, but its stock still traded below its $18 IPO price.