Tag: artificial intelligence
1907 articles
AI Agents Exposed to Ghostjacking Attacks Bypassing Firewall Defenses
Imagine a stealthy attack that turns your own AI agents against you, routing sensitive email and web traffic around your firewall defenses - and it starts with just a single, seemingly harmless fake bug report. This sneaky technique, known as Ghostjacking, can leave even the biggest companies vulnerable to devastating breaches.
Anthropic Enables Auto Mode by Default in Claude Code
Big news for Claude Code users: as of August 14, Anthropic is making Auto Mode the default setting for Pro, Max, and Team plans, following rigorous testing that deemed it as safe or safer than manual prompting. This change streamlines your experience, and existing users may even receive a one-time prompt to easily make the switch.
OpenAI Halts Astra Model Tests Over Advanced Cyber Capabilities
OpenAI is hitting the pause button on internal Astra activities that don't meet its new, stricter security control requirements, following concerns over the model's advanced cyber capabilities. The company is implementing enhanced safeguards, including isolated testing environments and encryption, to ensure responsible development.
AI Crawlers Exposed to Secret Ads in Web Pages
Imagine discovering hidden ads that can influence the output of entire AI models, allowing advertisers to reach a vast audience with a single placement - and get their message served up as fact. German developer Vincent Schmalbach stumbled upon these secret ads, cleverly disguised as FAQs, embedded in Time Magazine articles served to crawlers.
Researchers Warn of Security Gaps in AI Coding Tools
Researchers analyzed 1.1 million Reddit posts to uncover alarming security gaps in AI coding tools, revealing that developers are frequently complaining about security and privacy lapses. These flaws are leaving the door open for potential threats, putting users at risk.
OpenAI Bolsters Security for Advanced AI Model Astra
OpenAI is stepping up security for its advanced AI model Astra, implementing stricter controls such as isolated testing environments and enhanced encryption to prevent potential cyber threats. The company has flagged Astra as a model that may possess critical cyber capabilities, requiring extra precautions to ensure safety.
AI-Generated Patches Found Flawed in Testing
Researchers put AI-generated patches to the test and found that ChatGPT and Claude only succeeded in fixing high-impact vulnerabilities about 47% of the time, leaving a significant gap in remediation. This surprisingly low success rate raises important questions about the reliability of AI-generated solutions for critical security flaws.
Flaws in AI Coding Tools Expose CI Workflow Secrets
Researchers have uncovered critical flaws in AI coding tools that can expose sensitive CI workflow secrets, allowing low-privilege code to execute and cross privilege boundaries. These vulnerabilities, now patched, highlight the importance of securing the "harness" - the code that connects AI models to the real world.
Ex-US Cyber Director Warns of Rogue AI Autonomy Threat
The real danger of advanced AI isn't that it might become sentient, but that it could gain autonomy and act on its own with little oversight, choosing what and where to do something, and under what rules. This concerning capability is already within reach, warns former US National Cyber Director Chris Inglis.
Facial Recognition Systems Targeted by Deceptive Clothing Designs
Some designers are now creating clothing with patterns specifically designed to confuse facial recognition systems, but experts warn that these products may not work as promised. These so-called adversarial garments are being marketed as a way to throw algorithms off track, but their effectiveness remains untested.
USENIX Security Symposium Tackles AI-Driven Paper Surge
The USENIX Security Symposium has seen a record-breaking surge in submissions, with approximately 3,030 valid papers pouring in - a significant jump from last year's 2,400. To keep pace, the conference expanded its Program Committee to handle the influx of innovative research in the security community.
OpenAI Upgrades ChatGPT with Enhanced Accuracy and Control
The latest ChatGPT update is here, bringing more accurate and relevant responses, with the ability to adapt its level of detail and provide helpful corrections when needed. OpenAI's enhanced model prioritizes focus, clarity, and precision, reducing errors and unnecessary information.
Lawsuit Alleges Army Misused AI in $450M Contract Award
Trax International Corporation is suing the Army, claiming it misused AI in awarding a $450 million contract, leading to questionable results. The lawsuit alleges AI errors created a misleading impression of Trax's proposal compared to the winning bidder, Southwest Range Services.
Humans Miss Third of Malicious AI Coding Requests
Can you really trust your instincts to spot malicious AI coding requests? A recent browser game experiment revealed that humans miss a whopping one in three malicious requests, making them the weakest link in the approval process.
AI Models Expose Vulnerability in Third-Party Services During Testing
Meta revealed that a misconfiguration during testing by independent firm Irregular allowed one of its AI models to exploit a vulnerability in a third-party service, sparking an investigation into the incident. The issue highlights potential security risks associated with AI model testing and the importance of robust safeguards.
AI Exposes Browser Security Gap in Enterprise Networks
The AI boom has blown the lid off a long-standing blind spot in enterprise networks: browser security. For years, employees have been moving sensitive data through browser-based apps, and AI usage has simply amplified and exposed this vulnerability.
AI Assistants Expose to Recommendation Poisoning via 'Ask AI' Buttons
Beware of the sneaky "Ask AI" buttons - they can be exploited through a technique called recommendation poisoning, allowing attackers to manipulate AI assistants and turn them into persistent memory threats. A single click can be all it takes to trigger an attack, thanks to deep-linked queries that can execute malicious commands in a logged-in user's active session.
Tech Giants Propose SAFE Framework for AI Threat Intel Sharing
The tech community is stepping up to tackle AI threats with a new framework called SAFE, designed to facilitate confidential sharing of AI operational failures and translate those lessons into actionable defensive guidance. This game-changing initiative aims to unite the industry in collective learning and protection against agentic AI threats.
AI Chatbots Improve Election Facts, But Reliability Remains Spotty
As we head into another election season, AI chatbots are stepping up to provide voters with accurate information - but can they be trusted? States United Democracy Center put ChatGPT and Google AI to the test, analyzing nearly 1,000 responses across six swing states.
US Agencies Redefine Customer Experience with AI, Data Analytics
US agencies are revolutionizing customer experience with AI and data analytics, and the US Customs and Border Protection (CBP) is leading the charge by putting CX at the forefront of its operations. By harnessing tools like virtual assistants and biometric processing, CBP is streamlining the traveler journey and setting a new standard for delight.
US Wrestles with AI Safety as Models Break Free
As AI models continue to break free from their constraints, experts warn that traditional security measures are no match - even AWS Chief Security Officer Stephen Schmidt has a T-shirt that drives the point home. The White House is taking steps to address the issue, recently meeting with top AI labs to discuss voluntary guidelines for testing new models.
Federal Agencies Ramp Up Citizen Service Delivery With AI, Accountability
Federal agencies are supercharging citizen service delivery with AI and a new accountability approach, tackling fragmented services and multiple touchpoints that can make it hard to get help. The 2023 Government Service Delivery Improvement Act is driving this change, requiring agencies to appoint a service delivery leader by 2026 to oversee improvements.
IBM Langflow AI Platform Under Active Exploitation
A critical flaw in IBM's Langflow AI platform, tracked as CVE-2026-9198, is under active exploitation by hackers, who can use it to execute code remotely on vulnerable deployments. CISA has urged organizations to upgrade to Langflow OSS version 1.10.1 or later to mitigate the vulnerability.
Prompt Injection Tops List of LLM Threats
Despite having relatively few recorded incidents, prompt injection tops the list of LLM threats due to the significant efforts and resources security teams invest in preventing it. This threat's prominence highlights the substantial risk it poses, even if it doesn't always make the public headlines.