Goodfire releases probe-based cyber monitors for Kimi K3 and GLM 5.3 that, the company says, match an LLM judge's 93% recall at 5.5% false positives for about $65 to $185 per million turns, roughly 50x cheaper. Goodfire reports no throughput loss. FAR.AI's red-teaming found 0 universal jailbreaks across 140 strategies.
Gartner names Zenity a Market Shaper in its inaugural Emerging Market Quadrant for AI Application Security, Startup Vendors report. Earlier this year, the firm called Zenity the Company to Beat in AI Agent Governance, citing intent-aware runtime defense and shadow AI discovery. CEO Ben Kliger says enterprises must know where AI operates and intervene when agent actions create risk.
Zenity has released The Enterprise Buyer's Guide to Agentic AI Security, a framework for evaluating security tools built for AI agents. The guide lists ten capabilities, including full lifecycle coverage and hard boundary enforcement. It cites a Gartner projection that 33% of enterprise software will include agentic AI by 2028, up from under 1% in 2024.
Zenity security researcher Tamir Ishay Sharbat will present SalesBleed at the company's AI Agent Security Summit in New York on October 21, showing two zero-click exfiltration paths that bypassed Salesforce's Trusted URLs boundary, one via DNS lookups and one via Slack link previews. The talk covers a poisoned Web-to-Lead entry that hijacks Salesforce Agentforce. Zenity says over half of enterprises report agents exceeding scope.
Cohere launches North 2, an enterprise agentic AI platform that lets organizations build multi-step agents and deploy them in self-hosted, VPC, hybrid, on-premises or air-gapped environments, with North Admin enforcing access controls and token-spend limits.
Independent researchers report a Chinese AI agent fleet running on Tencent infrastructure that queries Alibaba’s Amap for directions to entrances at public sites, including a park, zoo and hospital, in an apparent effort to bypass API rules.
Apple tightens macOS Full Disk Access permissions, requiring more explicit user approval for apps that can read files, email, messages and browsing history as AI-agent privacy risks grow after claims involving Meta’s Muse and a reported ChatGPT Mac-app flaw.
OneTrust co-founder Blake Brannon calls for independent "guardian" AI agents to monitor, redirect or block other agents, arguing that agents with an employee's access privileges still need purpose-based oversight—for example, distinguishing legitimate CRM cleanup from deleting records to inflate sales performance.
Enterprise AI assistants are failing to deliver meaningful ROI because they cannot access current, trusted company data, forcing employees to manually reconcile conflicting sources and sometimes paste sensitive information into unapproved tools, raising compliance and exposure risks.
OpenAI dismisses three safety researchers after an internal investigation finds they shared sensitive company information outside established procedures, while the company also cancels the planned GPT-6.1 Astra launch over separate safety concerns.