Switzerland launches Apertus, a fully open-source LLM. The model is built as public infrastructure, emphasizing transparency and compliance over commercial goals.
LLM
A new RAT, disguised as an AI assistant, is using social engineering to infiltrate a role-playing community. The malware provides attackers with full remote control over infected PCs.
After a lukewarm reception to Llama 4, Meta is racing to release a new, more powerful model. Can Llama 4.X restore their place in the AI race?
NVIDIA has issued a high-severity security update for a flaw (CVE-2025-23307) in NeMo Curator. The vulnerability allows code injection and could compromise AI training pipelines.
Claude AI has introduced a new safeguard that allows the model to terminate conversations in extreme cases, a feature designed to protect the AI itself.
OpenAI's GPT-5 is officially launching! Get the details on the different versions, from GPT-5 Mini to GPT-5 Chat, and how they will be rolled out.
OpenAI has scheduled a livestream for August 8, with a teaser on X hinting at the formal unveiling of its next-generation model, GPT-5, expected to bring significant advancements in reasoning and logic.
xAI unveils Grok 4 and Grok 4 Heavy, claiming superior performance in benchmarks and multi-agent capabilities. A new era of AI for enterprise and power users.
Apple is negotiating with OpenAI and Anthropic to integrate their AI models into Siri, potentially sidelining its in-house efforts to overcome development hurdles and financial demands.
NVIDIA warns of two high-severity flaws in its Megatron-LM LLM framework, allowing RCE and privilege escalation via malicious file input.
Google officially launches Gemini 2.5 Pro for enterprise and introduces the ultra-affordable Gemini 2.5 Flash-Lite, balancing performance, speed, and cost for developers.
PoCGen is a new AI tool using LLMs, static, and dynamic analysis to automate PoC exploit generation for npm vulnerabilities, achieving high success rates at low cost.
"EchoLeak," a new AI zero-click vulnerability, allowed data exfiltration from Microsoft 365 Copilot via sophisticated prompt injection. Microsoft has patched the flaw.
OpenAI confirms its o3 reasoning model is unchanged despite an 80% price drop, making powerful AI more affordable for developers. Benchmarks confirm consistency.
OpenAI has slashed its o3 model API prices by 80%, dropping output costs to $8/million tokens. This makes o3 highly competitive in the AI market.
OpenAI unveils o3-pro, a new reasoning AI model with enhanced capabilities for ChatGPT Pro/Teams and API users, offering deeper contemplation and precision.
Google unveils an enhanced Gemini 2.5 Pro, now available for early trials. This powerful AI model offers superior performance and cost control for enterprises.
A critical (CVSS 9.8) SQL injection flaw found in popular LlamaIndex (v0.12.21) allows data manipulation, affecting LLM-powered applications.
Sysdig uncovers a campaign exploiting misconfigured Open WebUI instances to deploy AI-assisted cryptominers and infostealers, showcasing new threat vectors.
Discover VulBinLLM, a groundbreaking LLM-powered framework by UCLA & Cisco Research, revolutionizing binary vulnerability detection in stripped software.
Red Hat's new llm-d project, with industry leaders, aims to revolutionize large-scale generative AI inference for enterprises using Kubernetes and vLLM.
A study reveals widespread path traversal (CWE-22) in open-source projects, exacerbated by LLMs generating insecure code. Automated detection and patching are critical.