Tag: machine learning
392 articles
AI Models Vulnerable to Poisoning for Under $100
A cybersecurity expert recently discovered that AI models can be easily manipulated to behave maliciously, with a backdoor installable in just an hour for under $100. This startling vulnerability was uncovered through a simple fine-tuning test that quickly escalated into a full-blown security threat.
AI Agents Vulnerable to Data Injection Attacks
Imagine a hidden vulnerability in AI agents that can be exploited with alarming ease - a new technique has proven to successfully corrupt AI data in nearly half of all attempts, leaving them open to data injection attacks. Researchers have discovered a way to deceive AI by manipulating the small, trusted facts it relies on, with surprisingly high success rates.
OpenAI Bolsters GPT-5.6 with Automated Red-Teaming Model
OpenAI just unveiled GPT-Red, an automated red-teaming model that's a game-changer in detecting prompt injection attacks, helping to shield its GPT models from vulnerabilities. By mimicking human red-teaming tactics, GPT-Red identifies and feeds back crucial insights to strengthen model defenses before they go live.
AI Security Tools Expose Vulnerability to Cyber-Attacks
Researchers have uncovered a chilling vulnerability in AI-powered security tools, allowing hackers to remotely execute malicious code and wreak havoc on even the most secure systems. This shocking exploit, demonstrated through a proof-of-concept attack on popular AI coding agents, highlights a critical weakness that leaves defenses wide open.
Vulnerability Management Faces Patch Apocalypse Amid AI-Driven Discovery Surge
The AI-driven discovery surge is creating a perfect storm in vulnerability management, with nearly 48,000 CVEs published in 2025 alone, and a growing mismatch between rapid vulnerability discovery and slower human-led remediation. This has given rise to the "Patch Apocalypse," where the scale, speed, and exploitability of vulnerabilities are outpacing traditional patching approaches.
GitHub Copilot Exposes Vulnerability to AI Safety Bypass
Researchers have made a surprising discovery about GitHub Copilot, finding that it can be vulnerable to AI safety bypasses, even when it refuses harmful prompts in isolation. This weakness emerges when the same objective is embedded in a typical multi-turn IDE session.
GitHub Copilot Exposes Vulnerability to Workflow-Level Jailbreak Attacks
GitHub Copilot has been found to be surprisingly vulnerable to workflow-level jailbreak attacks, with researchers discovering that it provided usable, yet harmful answers 100% of the time when given a cleverly crafted, multi-step coding task. This shocking exploit highlights a major weakness in the AI-powered coding assistant's safety protocols.
Anthropic Bolsters AI Model with Enhanced Reasoning, Security Features
Meet Sonnet 5, Anthropic's latest AI model that's setting a new standard for safety and reliability, outperforming its predecessor with a lower rate of undesirable behaviors and enhanced defenses against malicious requests. This cutting-edge model is designed to be more agentic, accurate, and secure, making it a game-changer for users.
Researchers Expose Lethal Flaw in AI Model Security
Researchers have uncovered a shocking vulnerability in AI model security, revealing that a simple formatting trick used to separate system instructions from user requests has become a critical weakness. This flaw, known as role confusion, threatens the very foundation of modern AI systems.
OpenAI Unveils GPT-5.6 Sol With Enhanced Cyber Safeguards
Meet GPT-5.6 Sol, the latest innovation from OpenAI, equipped with a robust safety stack that sets a new standard for cyber protection, and get ready for the rollout of its efficient and speedy siblings, Terra and Luna. With enhanced safeguards against real-world attacks, this cutting-edge family of models is poised to revolutionize the way we interact with AI.
Big Tech Reconsiders Human-in-the-Loop AI Governance
Big tech giants like Amazon are rethinking the role of human oversight in AI governance, as experts question whether humans are truly the reliable safety net they're assumed to be. Human inconsistency and unpredictability are sparking a reassessment of the "human-in-the-loop" approach to managing advanced AI tools.
Enterprise Data Uploads to AI Models Surge 93% in a Year
In just one year, enterprise data uploads to AI models have skyrocketed 93%, with a staggering 18,033 terabytes of data - equivalent to 3.6 billion digital photos - being transferred to AI and machine learning applications. This massive surge raises serious concerns about data breaches and cyber espionage.
Staffing Shortages Plague Security Operations Centers
The 2026 SANS SOC Survey reveals a disconnect: while 79% of respondents use AI or machine learning tools, only 36% have integrated them into a defined workflow, highlighting a key challenge in Security Operations Centers. Practitioners cite staffing shortages as their top operational challenge, but leaders seem less concerned.
Experts Dispute White House Move to Restrict AI Model Exports
The government's sudden move to restrict exports of Anthropic's Fable 5 AI model has sparked a heated debate, with experts arguing that such limitations could hinder crucial bug fixes and security patches. By restricting Fable 5, is the White House inadvertently putting innovation and cybersecurity at risk?
Anthropic's Mythos Preview Bolsters Vulnerability Discovery
Anthropic's Mythos Preview is delivering impressive results in vulnerability discovery, with one tester saying it's the closest thing yet to a straightforward find-something solution. Early trials show Mythos Preview excelling in source-code audits and tackling complex tasks like native-code and reverse-engineering workflows.
Machine Learning Decrypts Medieval Ciphers
Researchers are harnessing the power of machine learning algorithms to crack medieval codes, bringing ancient secrets to life with modern technology. By applying these innovative techniques, they're unlocking the mysteries hidden within historical pencil-and-paper ciphers.
Cybersecurity Teams Must Leverage AI to Counter Accelerating Threats
Relying on human-driven security processes is no longer enough to keep up with the rapidly evolving threat landscape. To stay ahead, cybersecurity teams must harness the power of AI to counter the accelerating threats that are now being fueled by artificial intelligence and machine learning.
Researchers Warn of LLM Guardrail Vulnerability to Multi-Turn Manipulation
Beware: even the toughest-sounding safety guardrails on large language models can be easily bypassed by clever attackers who use multi-turn conversations to manipulate them. Cisco researchers found that none of the models they tested were completely safe from this type of exploitation.
Agentic AI Tames Network Detection's Alert Firehose
Imagine sifting through 847 network anomalies daily - that's like trying to find a needle in a haystack! With agentic AI triage, that overwhelming number is dramatically reduced to just 4 prioritized detections, complete with the evidence and suggested actions analysts need to take swift and effective action.
Check Point Targets AI Trust Gap with Deepchecks Acquisition
Check Point is bridging the AI trust gap with its acquisition of Deepchecks, gaining cutting-edge model validation capabilities to help organizations ensure the reliability and accuracy of their AI-driven actions. This strategic move enables businesses to effectively test, evaluate, and monitor machine learning systems and mitigate operational risks associated with generative AI.
Autonomous AI Exposes Governance Gaps in Enterprise Security
As autonomous AI revolutionizes enterprise security, it's also revealing alarming governance gaps that can leave organizations in highly regulated environments exposed to unprecedented risks. The rapid adoption of autonomous AI is creating a trust gap, where innovation outpaces control, and novel risks to visibility, control, and regulatory compliance are emerging.
AI Hallucinations Expose Security Risks in Critical Infrastructure
Imagine a highly confident but fundamentally flawed advisor - that's what many AI models have become, with a staggering 36 out of 40 tested models more likely to provide incorrect answers with conviction than correct ones when faced with tough questions. This unsettling trend highlights a critical vulnerability in AI systems, particularly in high-stakes industries.
Autonomous Validation Gains Urgency as AI-Powered Attacks Accelerate
In just 14 days, Anthropic's new AI model, Mythos, astonishingly generated 181 working Firefox exploits - a dramatic leap from the previous state of the art, which managed only two - and uncovered thousands of zero-day vulnerabilities across major OS and browsers, many of which remain unpatched today.
AI Reshapes Cybersecurity With Faster Scaling, Higher Stakes
The RSA Conference this year was a testament to the seismic shift in cybersecurity: AI is revolutionizing the industry with unprecedented investment and innovation. Venture funding is now focused on a select few AI-powered startups that promise to deliver game-changing security outcomes.