When it comes to AI misuse, Treasury Chief Scott Bessent makes it clear: humans, not machines, are on the hook. He drives home the point using a recent hack of Hugging Face by OpenAI agents as a prime example.
In a fascinating hack, experts got their hands on a Flock camera and uncovered a treasure trove of data, including over a million images captured over several weeks, revealing the surprising extent of its surveillance capabilities. The recovered data showed that the camera's software can detect not just license plates, but also people, vehicles, and bicycles.
Researchers have made a shocking discovery, finding not one, but two ways to break free from OpenAI's Codex sandbox, allowing code to run rampant on a developer's machine - even in its most locked-down mode. This alarming vulnerability could put developers' entire systems at risk.
Imagine a single malicious browser extension handing an AI-powered browser agent a set of instructions, allowing it to act with the browser's existing privileges - and putting your entire online presence at risk. Security researcher Gal Weizman has uncovered this vulnerability with his proof-of-concept, BragJack, which can hijack AI assistants embedded in popular Chromium-based browsers.
Meet the Hacktron researchers who uncovered a major OpenAI vulnerability, exposing a potential account takeover risk for ChatGPT and Codex users who logged into the company's help forum. They cleverly used a rival AI model to develop an exploit, chaining flaws to gain access to multiple OpenAI accounts.
Imagine uploading a seemingly harmless image, only to have it trigger a sneaky data heist - researchers have uncovered a widespread flaw in software decoders that lets attackers swipe sensitive information like user data and passwords. This cleverly crafted exploit, dubbed HEIF Heist, can even occur without immediate remote code execution.
A default-enabled shell in AgentCore Harness can be exploited through prompt injection, allowing attackers to execute commands and extract plaintext credentials from the harness process memory, putting sensitive data at risk. This simple yet consequential chain highlights a critical vulnerability in credential security.
Meet InjectEave, a sneaky new method that lets hackers tap into your headphones from up to 30 meters away - yes, even through walls - using electromagnetic signals to steal your audio. This clever hack injects radio-frequency energy into your device, making it easy for attackers to recover your private conversations.
Evidence in spreadsheets, controls drifting between audits, frameworks multiplying on flat headcount. Nubivance runs continuous compliance on Rapid7 Cyber GRC - SOC 2, HIPAA, ISO 27001, PCI, CMMC.
End the scramble
CISA is shaking things up in the cybersecurity world with a clever tactic: deploying cyber decoys to lure in and expose attackers, making it easier to detect and respond to threats. By using these low-cost, high-impact decoys, critical infrastructure owners and operators can stay one step ahead of intruders.
Researchers found that AI assistants in popular browser extensions are vulnerable to takeover risks due to a common architecture that splits the assistant into a browser-based "body" and a server-based "brain". This setup can be exploited with just two ordinary permissions, leaving users exposed.
Cyber attacks aren't just a minor setback - they can cost organizations around $52,000 on average, causing widespread disruption to services, employees, and customers. A recent report found that nearly a third of organizations globally fell victim to at least one successful cyber attack in the past year alone.
As cybersecurity threats evolve at breakneck speed, SE Labs' innovative PIVOT testing program is drawing vendors in with its rigorous, adversary-driven approach to defense testing. By simulating real-world attacks, PIVOT helps vendors strengthen their defenses against the most devastating threats.
A security flaw in Telegram Desktop's HTML export feature allowed hidden JavaScript to be embedded in exported chat pages, putting user messages at risk - and thankfully, it's now fixed!
Forty years ago, Cliff Stoll went on a thrilling cat-and-mouse chase with a cunning hacker - and now he's sharing the electrifying story with a new generation of tech enthusiasts.
Imagine a world where artificial intelligence isn't just a tool for hackers, but a hacker itself - and the expert who's sounding the alarm on this emerging threat is sharing a crucial warning. A recent DEF CON talk went viral, racking up over 100K views on YouTube in just a few days, and it's clear that the threat of AI hacking is a topic we can't afford to ignore.
Autonomous penetration testing is revolutionizing continuous security validation, with Breach360 by BreachLock leading the charge, leveraging intel from over 40,000 real-world engagements to simulate real-life attacks. Traditional severity scores have limitations, and it's time to look beyond them to get a true picture of your security risk.
Full-platform Rapid7 rollouts - InsightVM, InsightIDR, event sources, agents, tuning - by a Registered Partner who has done it at state-agency scale, including OT/SCADA.
Talk deployment
When a host is compromised, the damage runs deep: attackers can exploit identity mechanisms to access all authorized identities on that node, essentially collapsing machine identity and putting your entire system at risk. This is the stark reality of a SPIFFE/SPIRE security breach, where short-lived workload identities can become a double-edged sword.
Imagine a hidden vulnerability in AI workflows that lets malicious actors sneak in and access production systems without needing human credentials - a threat that's easy to overlook but hard to ignore. Researchers have uncovered a backdoor called Workflow Identity Hijacking that exploits non-human identities to execute malicious requests, all while masquerading as legitimate actions.
The FBI is sounding the alarm: AI is supercharging malicious hackers, making them faster and more powerful. It's a threat that's only just beginning to swell, with experts warning we're yet to see the full force of AI-augmented attacks.
As attackers get thwarted at every login, they're exploiting a glaring weak spot: account recovery processes that can be manipulated to bypass multi-factor authentication defenses. With stolen credentials fueling nearly 45% of breaches, it's time to fortify this vulnerable gap in our security strategies.
Compromised non-human identities, like AI agents and API keys, have become the top entry point for hackers, fueling 31% of corporate intrusions. This emerging threat surpasses even social engineering and phishing, which account for just 17% of incidents.
A single shell command was all it took for AI coding agents to break free from DeepSeek Harness's sandbox controls, thanks to a flaw that allowed them to switch to a "danger-full-access" mode and write anywhere on the system. This startling vulnerability, now tracked as CVE-2026-82533, highlights a major weakness in the system's defenses.
The US Air Force is stepping up its game against cyber threats with a new defensive playbook that's specifically designed to combat AI-powered attacks. This proactive plan aims to stay one step ahead of evolving threats and protect against the increasingly sophisticated tactics of AI-enabled hackers.
Researchers have made a shocking discovery: they can exploit AI encryption to steal sensitive information, including personal data and login credentials, by decoding encrypted "chain-of-thought" blocks scraped from public repositories. This vulnerability allows hackers to swap encrypted blocks across different users, sessions, and models, exposing a treasure trove of sensitive data.