Explore

Penetration Tests Reveal Severe AI Security Flaws Compared to Legacy Software

Penetration Tests Reveal Severe AI Security Flaws Compared to Legacy Software

Recent penetration tests have uncovered that AI-based systems exhibit a higher percentage of high-risk security flaws compared to legacy software. According to Cobalt's State of Pentesting Report, 32% of AI and large language model (LLM) vulnerabilities are rated as high risk, significantly higher than the 13% found in traditional enterprise security tests. These AI vulnerabilities also have the lowest resolution rate, with only 38% of high-risk issues being addressed. The report highlights the challenges in securing AI systems, which often introduce new attack surfaces due to their non-deterministic nature and integration with sensitive data.

Braintrust AI Platform Faces Security Breach, Urges API Key Rotation

Braintrust AI Platform Faces Security Breach, Urges API Key Rotation

Braintrust, a startup focused on AI model evaluation, has reported unauthorized access to one of its Amazon Web Services cloud accounts. The breach involved the compromise of secret API keys used by customers to connect to cloud-based AI models. While there is no evidence of a large-scale data breach, the company has advised customers to rotate their API keys as a precaution. An investigation is underway to determine the exact cause of the security incident, and access permissions have been restricted to contain the breach.

AI Security Risks Expose Vulnerabilities in Proprietary Software and Hardware

AI Security Risks Expose Vulnerabilities in Proprietary Software and Hardware

AI Security Risks Expose Vulnerabilities in Proprietary Software and Hardware

US Army and Tech Firms Collaborate on AI Cyber Defense Strategies

US Army and Tech Firms Collaborate on AI Cyber Defense Strategies

The US Army, in collaboration with major tech companies, is testing AI-driven cyber defense systems to enhance response times to cyberattacks. During a tabletop exercise at the Pentagon, known as AI TTX 2.0, military officials and executives from companies like Amazon Web Services, Google, Microsoft, and OpenAI explored how AI systems could outperform human operators in managing large-scale cyber threats. The exercise was organized by the Office of the Principal Cyber Advisor and included participation from US Cyber Command and US Army Cyber Command. The initiative aims to leverage AI's capabilities to detect and respond to cyber threats more efficiently, potentially transforming military cyber defense strategies.

Enterprise AI Faces Security Threat from Data Poisoning, Impacting Decision-Making

Enterprise AI Faces Security Threat from Data Poisoning, Impacting Decision-Making

Enterprise AI Faces Security Threat from Data Poisoning, Impacting Decision-Making

Claude Mythos AI Model Raises Security Concerns Among Tech Giants

Claude Mythos AI Model Raises Security Concerns Among Tech Giants

Anthropic's new AI model, Claude Mythos, has raised significant security concerns due to its ability to identify zero-day vulnerabilities in major software systems. The model, which is not being released publicly, has been shared with a select group of tech companies, including Google, Apple, and Microsoft, to address potential security threats. The initiative, known as Project Glasswing, aims to allow these companies to fix vulnerabilities before they can be exploited by malicious actors. The model's capabilities have prompted discussions about the need for industry-wide cooperation to prevent AI from being weaponized.

AI Startup Braintrust Confirms Security Breach, Urges Customers to Rotate API Keys

AI Startup Braintrust Confirms Security Breach, Urges Customers to Rotate API Keys

AI Startup Braintrust Confirms Security Breach, Urges Customers to Rotate API Keys

Braintrust Urges API Key Rotation Following Security Breach

Braintrust Urges API Key Rotation Following Security Breach

Braintrust Urges API Key Rotation Following Security Breach

Perplexity AI Expands Mac Tool Availability to All Users, Enhancing Workflow Management

Perplexity AI Expands Mac Tool Availability to All Users, Enhancing Workflow Management

Perplexity has announced the release of its Personal Computer tool for Mac devices, making it available to all Mac users. Previously, this tool was exclusive to Perplexity Max subscribers and required a waitlist. The app allows AI agents to access local files, applications, and connectors, as well as the web, to manage personal workflows securely. This approach is considered more secure than other AI solutions like OpenClaw, which posed security risks due to high permissions. Users can download the new Perplexity Mac app directly, but the Personal Computer feature requires a Pro or Max subscription. The app can work with local files, native Mac apps, and web tools, managing over 400 connectors in a secure environment. Additionally, the app can be integrated with Perplexity's AI-powered Comet web browser, allowing for web operations without direct connectors. Designed for always-on devices like Mac Mini, it allows remote access via iPhone to start tasks or approve requests.

Cyera Acquires Ryft to Enhance AI Security Capabilities

Cyera Acquires Ryft to Enhance AI Security Capabilities

Cyera Acquires Ryft to Enhance AI Security Capabilities

Australia regulator calls for urgent cybersecurity action to counter Mythos

Australia regulator calls for urgent cybersecurity action to counter Mythos

By Scott Murdoch SYDNEY, May 8 (Reuters) - Australia's corporate regulator has urged the country's financial sector to take urgent action on tackling potential cyber risks from frontier AI systems such as Mythos.

AI Adoption Outpaces Safety Policies, Exposing Organizations to Cyber Risks

AI Adoption Outpaces Safety Policies, Exposing Organizations to Cyber Risks

AI Adoption Outpaces Safety Policies, Exposing Organizations to Cyber Risks