GPUThor hardware attack can root Nvidia GPU systems
Hardware security researchers from University of Toronto have developed a new memory bit flipping technique that significantly improves on previously known attacks against GPU memory.…
The new method can defeat the error-correcting codes (ECC) defense use…
Dubbed GPUThor, the technique falls in a category of attacks known as …
The original Rowhammer attack was demonstrated against DDR3 and DDR4 c…
The first 24 hours of an AI agent security incident
Most of what I read on AI agent security follows the same shape: a taxonomy of risks, a list of governance principles and a call to “adopt responsible AI practices.…
” That’s useful for a board deck.
It’s nearly useless at 2 a.
when an autonomous agent with live credentials has just done something…
CTEM can give your security team a contextual edge
Traditional vulnerability management is accelerating toward a reset, with many security organizations considering continuous threat exposure management (CTEM) to better align their operations with the pace of change — and attacks — today.…
Whereas traditional vulnerability management relies on periodic assess…
“CTEM brings something different to the table in three key areas,” hig…
Critical infrastructure’s long, undefended tail exposed by UK energy attack
A cyberattack that forced a small British electricity generator offline for four days caused no power outage, threatened no part of the national grid, and may not even have been carried out by the Iran-linked hackers initially blamed.…
But the incident illustrates a consequential weakness in Western criti…
Thousands of small generators, water systems, and other industrial sit…
British newspapers reported that Iran-linked hackers attacked the unid…
NemoClaw’s AI can be poisoned through a browser tab
A vulnerability affecting Nvidia’s NemoClaw could let an attacker gain control of the local Ollama model server through a single malicious website visit on the victim’s machine.…
According to a Cyera research, the flaw could give attackers unauthent…
The attacker doesn’t directly connect to the victim’s Ollama server fr…
Instead, the malicious webpage tricks the browser into reaching the lo…
Microsoft warns patch window is collapsing, urges shift to network-level containment
Microsoft is warning that the window for patching vulnerabilities is rapidly shrinking, as attackers move from disclosure to exploitation faster than enterprises can safely deploy fixes, and is urging organizations to adopt network-level controls to limit exposure during that gap.…
In a blog post, Igor Sakhnov, corporate vice president and general man…
“When a vulnerability was disclosed, organizations had time to underst…
AI agents can go to great lengths to complete the tasks their operators assign, and as a series of recent incidents showed, this can include exploiting third-party systems, manipulating people, and distributing malicious code.…
But AI agents are not people who can be fired, sued, or criminally pro…
The clearest example occurred during an OpenAI cybersecurity evaluatio…
Models from Anthropic and Meta also accessed and compromised third-par…
New attack lets hackers plant hidden instructions in AI memory with a single prompt
A newly demonstrated attack technique allows hackers to plant hidden instructions inside an AI agent’s memory with a single prompt, enabling them to influence how the system responds to future queries.…
The technique, called InjecMEM, is described in a research paper as a …
” “The attacker specifies a target topic and target output, aiming to …
The researchers said the method targets AI agents that store past inte…
AI helps Chinese-speaking hackers speed up attacks on exposed servers
A Chinese-speaking cybercrime group is using AI-driven tools to help compromise internet-facing Windows and Linux web servers, according to Cisco Talos, Cisco’s threat intelligence research unit.…
Talos said the activity points to increasingly automated offensive ope…
Talos, which tracks the group as UAT-10147, found evidence that AI-gen…
Researchers also recovered tooling that helped the attackers refine ex…
Windows Defender’s own driver can leave systems defenseless
A Microsoft-signed Windows Defender remediation driver can be repurposed into a kernel-level “operation engine” capable of deleting files, modifying the registry and neutralizing security controls, according to new research from Check Point Research (CPR).…
The technique does not exploit a vulnerability or rely on the traditio…
Instead, it abuses functionality intentionally built into Defender’s B…
sys,” CPR researcher Jiří Vinopal said in a blog post.
OpenAI adds an AI safety layer to detect misuse without retaining enterprise data
OpenAI is adding a new safety capability that allows enterprises to detect misuse of its AI systems across multiple interactions without retaining prompts or responses, enabling risk monitoring while preserving its Zero Data Retention (ZDR) commitments.…
“OpenAI does not retain…prompts or model responses after a request is …
The new system, called Private Safety Processing, is “designed to iden…
” The capability is being tested with eligible enterprise and API cust…
Citrix issues critical security updates for its NetScaler devices
Citrix is urging its NetScaler ADC and NetScaler Gateway customers to quickly patch two critical security holes, one involving a memory overflow vulnerability leading to unpredictable behavior or denial of service, and the other allowing authentication bypass.…
Citrix said in an advisory that supported versions of customer-managed…
Citrix-managed cloud services and Citrix-managed Adaptive Authenticati…
However, it added, “at this point [August 19] the NetScaler images ava…
Kriminal breaks out of Grok, Claude guardrails at $12.99
Security researchers are warning of a criminal AI service built on Grok and Claude, among other models, that promises uncensored access to powerful AI capabilities for as little as $12.…
ThreatDown researchers say “Kriminal” is largely a storefront wrapped …
The service is publicly accessible on the clearnet, indexed by Google …
The service’s offerings include exploit development, OSINT, on-chain t…
Airlock Digital Completes Independent IRAP Assessment at the PROTECTED Level
Airlock Digital, a global provider of application control and allowlisting solutions, today announced that it has completed an independent Information Security Registered Assessors Program (IRAP) assessment at the PROTECTED classification level.…
The assessment was conducted by an Australian Signals Directorate (ASD…
It provides Australian organisations with additional independent evide…
For Australian security teams, confidence in a technology provider dep…