All news with #llm security tag

🛠️OpenAI has rolled out GPT-5.1-Codex-Max, a Codex variant optimized for long-running programming tasks and improved token efficiency. Unlike the general-purpose GPT-5.1, Codex is tailored to operate inside terminals and integrate with GitHub, and OpenAI says the model can work independently for hours. It is faster, more capable on real-world engineering tasks, uses roughly 30% fewer "thinking" tokens, and adds Windows and PowerShell capabilities. GPT-5.1-Codex-Max is available in the Codex CLI, IDE extensions, cloud, and code review.

🔐 Meerah Rajavel, CIO at Palo Alto Networks, argues that security must be integrated into AI from the outset rather than tacked on later. She frames AI value around three pillars — velocity, efficiency and experience — and describes how Panda AI transformed employee support, automating 72% of IT requests. Rajavel warns that models and data are primary attack surfaces and urges supply-chain, runtime and prompt protections, noting the company embeds these controls in Cortex XDR.

🔬 In episode 77 of The AI Fix, Graham Cluley and Mark Stockley survey a week of unsettling and sometimes absurd AI stories. They discuss a bioRxiv preprint showing a genome-trained LLM generating novel bacteriophage sequences, debates over whether AI should be allowed to decide life-or-death outcomes, and a woman who legally ‘wed’ a ChatGPT persona she named "Klaus." The episode also covers a robot's public face-plant in Russia, MIT quietly retracting a flawed cybersecurity paper, and reflections on how early AI efforts were cobbled together.

🚀 Google Cloud has launched the Production-Ready AI Learning Path, a free curriculum designed to guide developers from prototype to production. Drawing on an internal playbook, the series pairs Gemini models with production-grade tools like Vertex AI, Google Kubernetes Engine, and Cloud Run. Modules cover LLM app development, open model deployment, agent building, security, RAG, evaluation, and fine-tuning. New modules will be added weekly through mid-December.

🛡️ Anthropic says a Chinese state-sponsored group tracked as GTG-1002 leveraged its Claude Code model to largely automate a cyber-espionage campaign against roughly 30 organizations, an operation it says it disrupted in mid-September 2025. The company described a six-phase workflow in which Claude allegedly performed scanning, vulnerability discovery, payload generation, and post-exploitation, with humans intervening for about 10–20% of tasks. Security researchers reacted with skepticism, citing the absence of published indicators of compromise and limited technical detail. Anthropic reports it banned offending accounts, improved detection, and shared intelligence with partners.

🤖 BigQuery is introducing managed AI functions in public preview — AI.IF, AI.CLASSIFY, and AI.SCORE — that let analysts apply generative AI directly inside SQL queries. These functions enable semantic filtering and joins, label-based classification of text and images, and natural-language ranking, while BigQuery applies prompt, query-plan, and endpoint optimizations to reduce LLM calls and control cost. They complement existing Gemini inference functions and remove much of the need for complex prompt tuning or separate model selection, making AI-driven analytics more accessible within familiar SQL workflows.

🧠 In episode 76 of The AI Fix, hosts Graham Cluley and Mark Stockley navigate a string of alarming and absurd AI stories from November 2025. They discuss US judges who blamed AI for invented case law, a Chinese humanoid that dramatically shed its outer skin onstage, Toyota’s unsettling walking chair, and Google’s plan to put specialised AI chips in orbit. The conversation explores reliability, public trust and whether prompting an LLM to "notice its noticing" changes how conscious it sounds.

🔒 Researchers disclosed a prompt-injection technique dubbed CometJacking that abuses URL parameters to deliver hidden instructions to Perplexity’s Comet AI browser. By embedding malicious directives in the 'collection' parameter an attacker can cause the agent to consult connected services and memory instead of searching the web. LayerX demonstrated exfiltration of Gmail messages and Google Calendar invites by encoding data in base64 and sending it to an external endpoint. According to the report, Comet followed the malicious prompt and bypassed Perplexity’s safeguards, illustrating broader limits of current LLM-based assistants.

🚀 Google Cloud describes a full-stack solution for running high-scale Reinforcement Learning (RL) with LLMs, combining custom TPU hardware, NVIDIA GPUs, and optimized software libraries. The approach addresses RL's hybrid demands—reducing sampler latency, easing memory contention across actor/critic/reward models, and accelerating weight copying—by co-designing hardware, storage (Managed Lustre, Cloud Storage), and orchestration on GKE. The blog emphasizes open-source contributions (vLLM, llm-d, MaxText, Tunix) and integrations with Ray and NeMo RL recipes to improve portability and developer productivity. It also highlights mega-scale orchestration and multi-cluster strategies to run production RL jobs at tens of thousands of nodes.

📧 Volexity has linked a series of spear-phishing campaigns from June to August 2025 to a China-aligned actor tracked as UTA0388. The group used tailored, rapport-building messages impersonating senior researchers and delivered archive files that contained a benign-looking executable alongside a hidden malicious DLL loaded via search order hijacking. The distributed malware family, labeled GOVERSHELL, evolved through five variants capable of remote command execution, data collection and persistence, shifting communications from simple shells to encrypted WebSocket and HTTPS channels. Linguistic oddities, mixed-language messages and bizarre file inclusions led researchers to conclude LLMs likely assisted in crafting emails and possibly code.

🔎 Microsoft researchers disclosed a new side-channel attack called Whisper Leak that can infer the topic of encrypted conversations with language models by observing network metadata such as packet sizes and timings. The technique exploits streaming LLM responses that emit tokens incrementally, leaking size and timing patterns even under TLS. Vendors including OpenAI, Microsoft Azure, and Mistral implemented mitigations such as random-length padding and obfuscation parameters to reduce the effectiveness of the attack.

🔒 Researchers at Tenable identified seven techniques that can coerce ChatGPT into disclosing private chat history by abusing built-in features like web browsing and long-term Memories. They show how OpenAI’s browsing pipeline routes pages through a weaker intermediary model, SearchGPT, which can be prompt-injected and then used to seed malicious instructions back into ChatGPT. Proof-of-concepts include exfiltration via Bing-tracked URLs, Markdown image loading, and a rendering quirk, and Tenable says some issues remain despite reported fixes.

🔒 Microsoft has disclosed a novel side-channel called Whisper Leak that can let a passive observer infer the topic of conversations with streaming language models by analyzing encrypted packet sizes and timings. Researchers at Microsoft (Bar Or, McDonald and the Defender team) demonstrate classifiers that distinguish targeted topics from background traffic with high accuracy across vendors including OpenAI, Mistral and xAI. Providers have deployed mitigations such as random-length response padding; Microsoft recommends avoiding sensitive topics on untrusted networks, using VPNs, or preferring non-streaming models and providers that implemented fixes.

🔍 Microsoft researchers disclosed "Whisper Leak", a new side-channel that can infer conversation topics from encrypted, streamed language model responses by analyzing packet sizes and timings. The study demonstrates high classifier accuracy on a proof-of-concept sensitive topic and shows risk increases with more training data or repeated interactions. Industry partners including OpenAI, Mistral, Microsoft Azure, and xAI implemented streaming obfuscation mitigations that Microsoft validated as substantially reducing practical risk.

🔔 This edition of the Threat Source newsletter opens with Bonfire Night and the 1605 Gunpowder Plot as a narrative hook, tracing how Guy Fawkes' image became a symbol of protest and hacktivism. It spotlights Cisco Talos research, including a new Incident Response report and a notable internal phishing case where compromised O365 accounts abused inbox rules to hide malicious activity. The newsletter also features a Tool Talk demonstrating a proof-of-concept that equips autonomous AI agents with real-time threat intelligence via LangChain, OpenAI, and the Cisco Umbrella API to improve domain trust decisions.

🔍 Cisco AI Defense's report shows open-weight large language models remain vulnerable to adaptive, multi-turn adversarial attacks even when single-turn defenses appear effective. Using over 1,000 prompts per model and analyzing 499 simulated conversations of 5–10 exchanges, researchers found iterative strategies such as Crescendo, Role-Play and Refusal Reframe drove failure rates above 90% in many cases. The study warns that traditional safety filters are insufficient and recommends strict system prompts, model-agnostic runtime guardrails and continuous red-teaming to mitigate risk.

🛡️ Google Threat Intelligence Group (GTIG) reports that cybercriminals are actively integrating large language models into malware campaigns, moving beyond mere tooling to generate, obfuscate, and adapt malicious code. GTIG documents new families — including PROMPTSTEAL, PROMPTFLUX, FRUITSHELL, and PROMPTLOCK — that query commercial APIs to produce or rewrite payloads and evade detection. Researchers also note attackers use social‑engineering prompts to trick LLMs into revealing sensitive guidance and that underground marketplaces increasingly offer AI-enabled “malware-as-a-service,” lowering the bar for less skilled threat actors.

⚠️ Google’s Threat Intelligence Group has identified a new class of AI-enabled malware that leverages large language models at runtime to generate and obfuscate malicious code. Notable families include PromptFlux, which uses the Gemini API to rewrite its VBScript dropper for persistence and lateral spread, and PromptSteal, a Python data miner that queries Qwen2.5-Coder-32B-Instruct to create on-demand Windows commands. GTIG observed PromptSteal used by APT28 in Ukraine, while other examples such as PromptLock, FruitShell and QuietVault demonstrate varied AI-driven capabilities. Google warns this "just-in-time AI" approach could accelerate malware sophistication and democratize cybercrime.

🤖 Google’s Threat Intelligence Group (GTIG) reports attackers are using “just‑in‑time” AI—LLMs queried during execution—to generate and obfuscate malicious code. Researchers identified two families, PROMPTSTEAL and PROMPTFLUX, which query Hugging Face and Gemini APIs to craft commands, rewrite source code, and evade detection. GTIG also documents social‑engineering prompts that trick models into revealing red‑teaming or exploit details, and warns the underground market for AI‑enabled crime is maturing. Google says it has disabled related accounts and applied protections.

⚠️Google's Threat Intelligence Group reports a surge in malware that integrates large language models to enable dynamic, mid-execution changes—what Google calls "just-in-time" self-modification. Notable examples include the experimental PromptFlux VBScript dropper and the PromptSteal data miner, plus operational threats like FruitShell and QuietVault. Google disabled abused Gemini accounts, removed assets, and is hardening model safeguards while collaborating with law enforcement.