Explore the intersection of AI and cybersecurity. Stay informed on AI-driven security trends, tools, and threats in the ever-evolving digital landscape.
Search across headline titles and summaries.
Background for this topic.
Artificial intelligence (AI) describes computer systems that perform tasks such as recognizing patterns, making predictions, understanding language, or generating content. In security reporting, the term commonly includes machine-learning models used for detection and analysis, as well as generative AI applications that produce text, code, images, or other outputs.
AI can help analyze security telemetry, prioritize vulnerabilities, and support investigations, but its outputs can be wrong or manipulated. Important attack surfaces include prompt injection that steers an application into unintended actions, sensitive data being exposed through prompts or model outputs, and excessive permissions granted to AI systems that use external tools. Models can also be degraded by poisoned training data or evaded with carefully crafted inputs. Practitioners should protect training and operational data, limit model access and tool permissions, test for adversarial behavior, and require appropriate human validation before high-impact decisions.
Model Appears to Be a Way Station on the Road to Something GreaterOpenAI on Thursday released its latest generative AI model, but don't call it the next big thing just yet. More thoughtful, persuasive and emotionally intelligent, GPT-4.5 aims to feel less like an algorithm and more like a conversation partner.
LLMjacking operation leveraged illicit access GenAI services to produce explicit celebrity images and other harmful content, Microsoft's digital crimes unit says.
Law enforcement agencies from 19 countries have arrested 25 suspects linked to a criminal ring that was distributing child sexual abuse material (CSAM) generated using artificial intelligence (AI). [...]
Microsoft on Thursday unmasked four of the individuals that it said were behind an Azure Abuse Enterprise scheme that involves leveraging unauthorized access to generative artificial intelligence (GenAI) services in order to produce offensive and harmful content
A dataset used to train large language models (LLMs) has been found to contain nearly 12,000 live secrets, which allow for successful authentication
Crew helped lowlifes generate X-rated celeb deepfakes using Redmond's OpenAI-powered cloud – claim Microsoft has named four of the ten people it is suing for allegedly snatching Azure cloud credentials and developing tools to bypass safety guardrails in its generative AI services – ultimately to generate deepfake smut videos of celebrities and others.…
Series A Investment Expands AI-Driven Cybersecurity and Threat DeflectionMimic got $50 million in Series A funding to expand its ransomware defense solutions. Backed by Google Ventures and Menlo Ventures, the company will enhance AI-driven threat detection, automate security for proprietary apps, and grow internationally to protect enterprises from ransomware attacks.
Approach Aims to Scale AI Models by Making them Smarter Instead of BiggerScaling AI models in size has hit a plateau. One possible solution to is time test compute, which dynamically allocates extra computational resources during inference - or the thinking phase - to refine answers. Test-time compute lets AI models allocate resources based on the problem's complexity.
Also, AI Video Mocking Trump and Musk Disrupts HUD OfficesThis week, a U.S. Army soldier pleaded guilty, an AI video displayed to federal workers mocked Donald Trump and Elon Musk, a Saudi firm hit by ransomware, a new North Korean scam, hackers targeted Ukrainian notaries, CISA flagged two flaws, a botnet targeted Microsoft 365 and unpatched Ivanti VPNs.
Data poisoning represents the next big existential cybersecurity threat — unless organizations can ensure their AI systems are safe and trustworthy.
Microsoft has named multiple threat actors part of a cybercrime gang accused of developing malicious tools capable of bypassing generative AI guardrails to generate celebrity deepfakes and other illicit content. [...]
DeepSeek's Reasoning AI and Inference Scaling Driving Massive Demand for ComputeAI’s shift to reasoning models like DeepSeek is pushing compute demand to unprecedented levels, says NVIDIA CEO Jensen Huang. Blackwell chips are designed to handle inference scaling, where complex models consume 100x more compute than traditional AI. He predicts strong data center growth in 2025.
Model was fine-tuned to write vulnerable software – then suggested enslaving humanity Computer scientists have found that fine-tuning notionally safe large language models to do one thing badly can negatively impact the AI’s output across a range of topics.…
Javelin's Jennifer Pitt on AI-Powered Detection of Authorized Payment FraudMachine learning helps banks detect APP fraud by analyzing large transaction datasets faster. These AI models operate in near real-time and can distinguish legitimate transactions from fraudulent ones by spotting anomalies, said Jennifer Pitt, senior analyst at Javelin Strategy & Research.
Claude 3.7 Sonnet Can Give Rapid or Deliberate, Complex Answers to PromptsAnthropic introduced a new artificial intelligence model designed to adapt its reasoning time based on user preferences. Marketed as the industry's first "hybrid AI reasoning model," Claude 3.7 Sonnet aims to deliver rapid responses and more deliberate, complex answers to prompts.
99% of organizations report API-related security issues, highlighting risks from API growth
OpenAI's newest model, GPT-4.5, is coming sooner than we expected. A new reference has been spotted on ChatGPT's Android app that points to a model called "GPT-4.5 research preview," but it looks like it will initially be limited to those with a Pro subscription. [...]