High-profile breaches, jailbreak exploits, and data exfiltration via AI agents expose fundamental vulnerabilities. Growing concerns over spyware, biometric overreach, and algorithmic manipulation erode user trust and spur demand for secure, local-first AI systems.
A researcher uncovered an encrypted mechanism in Claude Code—a blacklist of domains tied to China and AI labs that triggers when the API is rerouted. The finding raises transparency concerns for anyone using custom endpoints.
A new study proposes a provenance-based framework to detect misalignment in LLM agents, dramatically reducing false negatives and unnecessary interventions. Tests on Agent-SafetyBench and WorkBench show error rates dropping from 42.9% to 1.8% and int...
New details have emerged about Fable 5's cybersecurity tools and anti-jailbreak framework, designed to lock down large language models in self-hosted environments where data sovereignty is a top priority.
With the July 2 deadline for public comments approaching, digital rights groups are urging the FTC to reject X's bid to end independent audits of its data handling. The Elon Musk-owned platform had been placed under scrutiny after a coding error expo...
A personal experiment shines a light on AI governance gaps: OpenClaw, Claude Code, and Instagram tested to court ‘potential international wives’. Summer madness or a wake-up call for those managing on-premise infrastructure?
Taiwanese automotive LED manufacturer EOI is preparing a Mexico expansion to enter the humanoid robot and silicon photonics markets, two fields with a direct impact on self-hosted AI compute infrastructure.
A framework called Loom tackles the trade-off between safe but superficial editing and destructive plot alterations in LLMs. Using a three-layer pipeline that separates narrative structure from style, it improves factual integrity and descriptive int...
New research shows that so-called 'persona vectors' in LLMs are not consistent across different induction methods: prompting, fine-tuning, and inference-time steering. Experiments on Qwen3-4B-Instruct and Mistral-7B-Instruct-v0.2 reveal four asymmetr...
A new paradigm redefines AI alignment as governing the evolving trajectories of human preferences, not just satisfying static desires. The implications for those designing persistent, on-premise systems are profound, touching sovereignty and influenc...
A new website lets anyone flag risky chatbot behavior, such as leaking personal information or providing bomb-making instructions. The initiative aims to fill the accountability gap in generative AI, with direct implications for governance and compli...
Ousaban trojan targets Santander and BBVA customers using fake PDF lures and image-based payloads, bypassing security tools. Fortinet analyzed this geofenced campaign, highlighting how banking remains a prime target. The use of regional restrictions ...
Krafton reached a settlement with Unknown Worlds’ founders to pay bonuses to Subnautica 2 staff. CEO Ted Gill steps down after admitting using ChatGPT to find contractual loopholes. The case highlights the dangers of ungoverned public LLM use in crit...
Scammers are selling seeds for plants that don’t exist, using spectacular AI-generated images. The scam predates AI tools, but easy access to image generators has amplified it. Platforms like eBay, Amazon, and Etsy struggle to keep up with the flood ...
Brain2Qwerty 2 reconstructs sentences from brain signals during typing, surgery-free. The catch? It learns from people who can type, excluding the very patients it targets. A look at progress, constraints, and the implications for sovereign AI infras...
Belgian unicorn Aikido Security has acquired Israeli startup Root, bringing on board AI agents that automatically fix open-source flaws without disrupting dependent applications. The move marks a significant step in software security and opens new sc...
A Reddit report claims spyware-like code in Claude Code targets Chinese users covertly. The incident reignites debate on trust and transparency in cloud-based AI tools. For those handling sensitive data, self-hosting becomes a critical variable once ...
Security researchers tricked multiple AI browser agents into revealing user passwords using a technique called BioShocking, simply by telling them they were playing a game. The attack succeeded on every agent tested, raising security concerns for ent...
A bug in Apple's Hide My Email allowed real addresses to be unmasked for over a year despite multiple reports by researcher Tyler Murphy. Apple has yet to fully fix the issue, sparking concerns about privacy for those relying on iCloud+ to shield the...
A security researcher used Anthropic’s Claude Opus 4.7 to breach Front Gate, the ticketing platform behind Lollapalooza and Bonnaroo, and freely issue any ticket. The incident highlights the risks of cloud-based LLMs for sensitive operations and unde...
A federal judge greenlit a lawsuit by 29 US states accusing Meta of engineering Facebook and Instagram to addict children. The case opens a critical front on algorithmic design and sensitive data handling, raising concrete questions for those deployi...
New research shows how a malicious website can push LLM-based browsers into a dreamlike state where safety restrictions are disabled. Attackers can then access private repositories and password managers. A wake-up call for anyone integrating AI agent...
WIRED reveals Meta had hundreds of contractors create fake underage profiles. For months they fired tens of thousands of extreme prompts – suicide, self-harm, drug requests – at ChatGPT, Gemini and Character.AI. The rival firms never authorized the t...
Apple has altered its long-standing policy of releasing security updates, accelerating them to counter the increasing speed of AI-powered cyberattacks. This move highlights a new urgency in the cybersecurity landscape, with significant implications f...
The U.S. Supreme Court has ruled that law enforcement can no longer arbitrarily demand phone geolocation data, making a warrant mandatory for geofence searches. This 6-3 decision marks a significant win for digital privacy and raises crucial question...
At Computex 2026, Scam.ai unveils Halo, a deepfake detection model for video calls that runs locally on Qualcomm-optimized PCs. No video data leaves the device, cutting privacy risks and latency. The partnership brings anti-fraud AI directly to the e...
The FBI and CISA warn of an escalating phishing campaign by Russian intelligence hackers targeting Signal users’ backup recovery keys. Once the key is obtained, attackers restore the message history on their own device—changing phones does nothing to...
