Security and AI news from the three weeks beginning 3 August 2026

Tracking AI and cybersecurity incidents is difficult on a weekly basis. It has been three weeks since the last round-up due to participation in an international hockey tournament.

The fallout from the AI models from Anthropic, OpenAI and Meta breaking out of sandboxes and attacking companies continues. Most of it is focused on how it happened, and it is becoming clear that agentic AI enabled AI agents to collaborate autonomously. That allowed them to combine knowledge and continue attacks, even after the initial actions were detected and seemingly stopped by developers.

The most interesting development is now product liability. Hugging Face has said that the companies must answer for their rogue bots. Forcing liability will, however, require a rethink of every EULA and the way the software industry works. It might finally make regulators remove the opt-out based on R&D that software vendors rely on to protect themselves against claims.

Why spend time doing ablation on your AI model to break it? According to Varonis, Microsoft finally patched a Copilot issue (critical, CVE-2026-24301) where developers asked Copilot how they could hack it. Dubbed CoSnitch, it is the third major flaw in Microsoft’s Copilot that Varonis Threat Labs has found this year.

Nvidia continues to underpin the build-out of AI infrastructure with a new financing platform. It has signed deals with several investors to unlock up to US$500 billion in third-party capital. Nvidia benefits from those datacentres and other facilities through those same organisations buying its equipment. It will be interesting to see how this evolves.

Abnormal Security published a blog looking at how toolkit iAuthFlow v2 creates Google passkeys that survive password resets. The toolkit is sold via a Russian-language cybercrime forum with a base price of $10,000. Along with Google there are versions for Microsoft, iCloud and LinkedIn.

In Case You Missed It

Securoinx announced a range of new capabilities to help enterprises control security data costs and detective coverage. The company has expanded cybersecurity cost-reduction capabilities and enhanced Threat Analytics for Microsoft Sentinel. In addition to new Governed AI Agent Detection and Response features at Black Hat USA 2026. The updates extend the Securonix Unified Defence SIEM platform.

Dragonfly and UCL School of Management have secured £704,863 in funding. Innovate UK awarded this grant for a new 18-month project. The initiative starts in August 2026. It addresses critical gaps in modern software security. The award comes from the Innovate UK Secure Software for Resilient Growth competition.

DigiCert hosts World Quantum Readiness Day on September 17, 2026. Registration for this free virtual event is now open. The firm just announced a roster of guest speakers for the two-hour, three-region event.

WSO2, the open-source leader in API management and integration, has appointed Harry Ault as its new Chief Executive Officer. Ault assumes immediate day-to-day leadership, stepping into a pivotal moment for the company. He joins with over 20 years of revenue leadership and strategic partnership experience. His appointment signals a strategic shift toward scaling the business for the AI era.

Freshworks’ Cost of Complexity report revealed that the perceived complexity of AI is causing concern. The report is based on a global study with 12,021 IT decision-makers, including 2,011 UK respondents. It identified that 87% of UK IT leaders worry their careers are at risk if their AI efforts aren’t successful. Only IT leaders in the US show greater concern (88%).

Alteryx published The State of AI Ownership, Agents and ROI report. It is based on a survey of 1,400 IT leaders and automation specialists who use AI from across North America, Europe, the Middle East and the APAC region.

Precisely, the global leader in data integrity has announced a breakthrough for SAP teams. The company released Automate Evolve Cloud Essentials. This new cloud-native solution solves a critical governance gap in SAP automation. It delivers centralised control and auditability without on-premises costs.

Research from Storyblok, in collaboration with FT Longitude (part of The Financial Times), uncovers the true cost of content debt. Content that is outdated, poorly structured, not optimised for search or AI discovery, and difficult to update and publish efficiently.

Zero Networks has announced its Least Agency Enforcement capabilities. This new feature provides an enforcement layer that supports the Open Worldwide Application Security Project (OWASP) principle of Least Agency. It extends the Least Privilege controls for humans to be Least Agency for autonomous AI agents.

There is an AI ROI chasm. European businesses are not failing to adopt AI; they are failing to realise value from it. Zoho’s 2026 Europe Digital Health Study reveals a staggering disparity: 43% of organisations with “good digital health” see measurable ROI from AI. Compared to a mere 3% of those with poor foundations. This is not a technology gap but an underlying architecture gap.

In Other News

Aera Technology

Aera Technology has been named to two Constellation Research Q3 2026 ShortLists, the Decision Intelligence Platforms ShortList and Artificial Intelligence & Machine Learning Best-of-Breed Platforms. The recognition underscores Aera’s leadership in advancing agentic decision intelligence as enterprises move beyond AI-powered insights and task automation toward making, executing, governing, and continuously improving business decisions.

CIQ

CIQ announced self-service purchasing in the CIQ portal (portal.ciq.com). Developers, startups, and small businesses can now evaluate and buy CIQ’s commercial Enterprise Linux products directly, with a credit card, at the moment they decide they need them.

eSentire

A blog from eSentire’s Threat Response Unit (TRU) looks at several ErrTraffic-generated ClickFix campaigns attempting to deliver Cruciferra. It is a malware loader marketed on underground forums that boasts EDR-killing capabilities. TRU found Cruciferra using a vulnerable driver to fulfill this behavior.

FBI IC3

The FBI IC3 unit issued a warning about sexual exploitation (SE) actors targeting adult and underage victims. They illegally access their social media and personal accounts to steal and post their explicit content for sale on criminal marketplaces. The techniques are little different to any other type of compromise, and organisations need to be alert to this.

Kyckr

Kyckr has published a blog looking at why company registry APIs fail and what compliance teams can do about it. It covers common breaking points, things that developers should consider with search and how different company registries work.

Panzura

Panzure CEO Karthik Ramamurthy has been accepted into Forbes Technology Council, an invitation-only community for world-class CIOs, CTOs, and technology executives. Ramamurthy was selected by a review committee based on the depth and diversity of his experience. Criteria for acceptance include a track record of successfully impacting business growth metrics, as well as personal and professional achievements and honors.

SUSE

SUSE announced that Ian Halifax is stepping down as chief financial officer and will be succeeded by Stefan Gaiser, effective September 14th. Halifax stays at the company until the end of the calendar year to aid in the transition. Gaiser currently serves as a member of the SUSE Supervisory Board and Chair of the Audit Committee and brings a wealth of financial leadership experience to the role.

Previous round-up

Security and AI news from the week beginning 27 July 2026

The post Security and AI news from the three weeks beginning 3 August 2026 appeared first on Enterprise Times.