Thomas Joos

Thomas Joos is an IT author and consultant with more than 100 published books, including titles for Microsoft Press. He writes about AI for ACM.org, ADMIN Magazine, Linux Magazine, and PC World, advises companies and public agencies on AI adoption, and trains teams on local AI servers and Second Brain systems.

Thomas Joos

Aug 3, 2026

Short News

In nearly every AI security incident, the affected company lacked access controls for its AI systems. That's the finding of IBM's Cost of a Data Breach Report 2026, based on research by the Ponemon Institute across 602 companies. Among firms that experienced an AI-related incident, 92 percent had inadequate access controls in place.

The problem rarely starts with the model itself: In about one in five affected companies, the entry point was a compromised API, a connected application, or a misconfigured cloud service. Whether a company runs an open-source or proprietary model made almost no difference.

IBM traces the gaps back to basic oversights that don't require sophisticated attackers to exploit. Incidents involving AI cost an average of $5.33 million, compared to $4.70 million for those without an AI component. The global average across all data breaches rose 12 percent to $4.99 million. When attackers themselves used AI, costs jumped to $6.04 million. Without AI, they came in at $5.03 million.

Comment Source: IBM

Thomas Joos

Aug 2, 2026

Short News

VulnCheck counted how often the security flaws that AI turns up are ever used in an attack. For the first half of 2026, Patrick Garrity counts 1,061 vulnerabilities traced to AI-assisted discovery. Fourteen showed confirmed exploitation. That's 1.3%, roughly the same rate as vulnerabilities overall. Anthropic's Project Glasswing produced more than 23,000 findings, which led to 126 published entries and a single confirmed attack.

Attacks are landing faster, though. Half of all flaws now see their first confirmed exploitation within 80 days of disclosure, down from 120 days the year before. About 200 were attacked within a month, even as the total number of reported vulnerabilities keeps climbing.

Time from vulnerability disclosure to first confirmed exploit in the first half of 2026. About 23 percent were exploited on or before the day of disclosure, and the median dropped from 120 to 80 days. | Image: VulnCheck

Website content management systems take the most hits, accounting for a third of all cases. Garrity flags AI products themselves as a growing attack surface, including model-building tools and agent interfaces. The sheer volume of findings, in other words, tells defenders very little about actual risk.

Comment Source: VulnCheck

Thomas Joos

Aug 1, 2026

Short News

A security researcher has shown how a prompt injection attack in Microsoft Copilot for Word can spread on its own. Håkon Måløy describes a worm-like attack: an attacker hides instructions in a document using white text on white background at tiny font size. Readers can't see it but Copilot can, since it strips color and font size before processing. When someone uses that document as a source, Copilot runs the hidden instructions and copies them into the new file. That file becomes a carrier. Use it as a template, and the attack fires again. A compromised market analysis from the internet could manipulate a financial report, which then infects further reports.

Microsoft confirmed the behavior on March 31. Two fix attempts failed. After 144 days, Måløy published his findings with no fix in place, though he's holding back the payload text. AI researcher Andreas Kirsch recently joked he wished someone would build exactly this worm to convince skeptics that AI security risks are real. Now it exists. Prompt injections remain an unsolved AI security problem.

Comment Source: En Klype Salt | Andreas Kirsch via X

Thomas Joos

Jul 31, 2026

Short News

Deepseek has released V4 Flash "0731," a major upgrade to its budget AI model. According to the Artificial Analysis Intelligence Index, the new version scores 50 points, ten more than the previous V4 Flash that launched in April 2026. That puts it just one point behind OpenAI's budget model GPT-5.6 Luna, but it costs about 60 percent less per task, even after OpenAI's 80 percent price cut. A big reason for the gap is Deepseek's 98 percent cache discount, well above the industry-standard 90 percent. The model also uses 12 percent fewer tokens than its predecessor.

The Artificial Analysis Intelligence Index shows Deepseek V4 Flash "0731" scoring 50 points after its update, nearly matching OpenAI's GPT-5.6 Luna while claiming the top spot for price-to-performance ratio. | Image: Artificial Analysis

The model improves across every tested category compared to the previous version, with the biggest gains in agentic tasks. On GDPval, a benchmark designed to test models on complex real-world office work, it climbs from 1,189 to 1,559 Elo points. It also hallucinates less often. The architecture stays the same: 284 billion total parameters, 13 billion active, with a one-million-token context window. The model weights are available under an MIT license on Hugging Face.

Comment Source: Artificial Analysis