All news with #research tag

🔍 Steven Murdoch uncovered that U.S. military satellites have been broadcasting hidden codes via public GPS for nearly two decades, effectively turning each satellite into a covert distribution channel. He identified synchronized transmissions across all 31 operational satellites on May 26, 2011, matching the rollout timeline of the military’s Over-the-Air Distribution (OTAD) and Over-the-Air Rekeying (OTAR) systems. This mechanism allowed remote rekeying of military GPS receivers, replacing manual key distribution.

🧩 Researchers at the University of Toronto built and tested a proof-of-concept self-replicating worm driven by a locally hosted open-weight large language model. In isolated experiments on a deliberately vulnerable 33-host network, the agent identified dozens of vulnerabilities, gained elevated access across most targeted hosts, and autonomously replicated to a majority of the network without using any commercial AI API. The team highlights how runtime reasoning and ingestion of fresh advisories break single-CVE patching assumptions and argues containment must focus on host and network controls rather than vendor API measures.

🛡️ Researchers at Graz University of Technology describe FROST, a browser-based timing attack that uses the Origin Private File System (OPFS) to infer which sites a user visits and which apps they open. The exploit runs purely in JavaScript, requires no native code or permissions, and sharpens timer resolution via cross-origin isolation. On macOS it achieved high fingerprinting accuracy, while mitigations remain limited and browser vendors have not implemented firm fixes.

🔒 Researchers have demonstrated a prototype AI-powered internet worm that autonomously propagates and carries its own local LLM to run on compromised machines. The prototype echoes early theoretical concepts of self-replicating code and shows how generative models can be embedded into malware to extend functionality. This proof-of-concept highlights evolving threats and the need for updated defensive strategies and policy responses.

🔍 Security researchers disclosed a flaw in default HTTP/2 configurations that enables a denial-of-service technique dubbed the "HTTP/2 Bomb." The issue abuses HPACK header compression and flow-control behavior to force excessive memory allocations and hold them, impacting servers such as nginx, Apache HTTP Server, Microsoft IIS, Envoy, and Cloudflare’s Pingora. Patches have been released for several implementations, and mitigations include disabling HTTP/2 or enforcing header count caps.

🧭 The post considers how historical plaintext-hiding techniques, traditionally done by hand, created patterns such as short key phrases that made ciphers vulnerable to statistical analysis. It argues that modern AI and LLMs, being fundamentally statistical models with some randomness, can exploit ciphertext statistics to reconstruct plaintext. The author notes this capability does not automatically make decryption trivial, but highlights the potential for AI to invert statistical patterns in encrypted text.

🛡️ A security researcher published exploit code for a Visual Studio Code zero-day that enables attackers to steal GitHub OAuth tokens by tricking users into clicking a link. The flaw abuses VS Code's sandboxed webview message-passing to run JavaScript that simulates keypresses, installs a malicious extension, and exfiltrates tokens sent to github.dev. The vulnerability is unpatched and unassigned a CVE; users can mitigate risk by clearing cookies and site data for github.dev to force reauthentication prompts.

🔍 Kaspersky analyzed public Wi‑Fi across Mexico City, Guadalajara, and Monterrey ahead of the 2026 World Cup. The team wardrove to log 84,500 signals and 69,500 unique SSIDs, finding about 82% use WPA2/WPA3 but over 10% are unsecured. WPS was enabled on roughly 45% of access points, often even when WPA2/WPA3 was in use, increasing attack risk. The report also warns of other travel threats like malicious QR codes, public USB chargers, NFC/Bluetooth exploits, and evil‑twin networks. Kaspersky recommends using cellular data or an eSIM and a VPN to stay safe when connecting to public networks.

🧭 At Infosecurity Europe 2026, OWASP will unveil the Agentic Research Council to better align fast‑moving agentic AI capabilities with security research and operational practice. Launched from the GenAI Security Project’s Agentic Security Initiative, the council will prioritize a public research pipeline, convene working groups and connect academic outputs to deployable mitigations. The initiative aims to accelerate runtime‑focused defenses against multi‑agent threats.

🔬 Researchers at ETH Zurich report creating a device that generates provably perfect random numbers using two superconducting qubits, a 30-meter microwave guide, and specialized software. The setup leverages quantum entanglement and an amplification algorithm to remove bias common in classical and many quantum sources. The team says the output can be certified and could serve cryptographic key generation or public randomness services.

📡 Researchers show WiFi signals can reveal people and environments by analyzing how radio waves reflect, scatter, and absorb compared with expected patterns. WiFi sensing uses these variations to infer spatial structure and presence, effectively creating an image of surroundings and occupants. Thorsten Strufe of KIT explains it functions like a camera, but with radio waves instead of light, enabling recognition through signal propagation analysis.

🔬 At Google, A/B experimentation extends beyond UI tweaks to critical infrastructure components like kernels, memory allocators, and schedulers. They run machine-level experiments on representative 1% subsets of the fleet to avoid selection bias and capture system-wide effects across colocated workloads. The framework enforces binary hermeticity and a strict two-step rollout so experiments can be activated and rolled back safely. Performance is assessed using application-defined productivity metrics, machine counters, and reliability signals.

🔍 Researchers from Xidian University describe a new image-based prompt injection called CrossMPI that uses near-imperceptible pixel perturbations to alter how large vision-language models interpret both visual and textual inputs. The technique targets intermediate multimodal fusion layers rather than final outputs, misleading LVLMs without modifying text prompts. Tests show strong black-box transferability and high success rates across several open-source models, while common defenses reduce but do not fully eliminate the threat.

🔒 On day one of Pwn2Own Berlin 2026 researchers earned $523,000 exploiting 24 unique zero-days, led by Orange Tsai, who collected $175,000 after chaining four logic flaws to escape the Microsoft Edge sandbox. Windows 11 was rooted three times for new privilege-escalation bugs, and Valentina Palmiotti secured payouts for Red Hat Workstations and an NVIDIA Container Toolkit flaw. The event focuses on enterprise and AI-targeted technologies.

🔧 At Cloudflare we encountered severe query slowdowns after changing partitioning for a large ClickHouse table to support per-namespace retention; the migration aimed to enable tenant-specific TTLs without thousands of tables. Usual metrics (I/O, memory, rows scanned, parts read) looked normal, but flame graphs exposed heavy lock contention in query planning and costly copies of a giant parts vector. We implemented shared locks, a shared cached parts view, and a binary-search-based prune on the partition key to avoid linear scans. These patches dramatically reduced SELECT latency and were contributed upstream.

🔍 In a RSA 2025 conversation, Allie Mellen, author of Code War, frames modern cyber conflict through historical doctrine, showing how nations' distinct strategies shape attacks and espionage. She cautions that attribution based solely on technical signals is insufficient because actors can forge signatures and deploy false flags, so motive and context matter. Mellen warns that AI will make attacks faster and more adaptive, and urges defenders to strengthen fundamentals and adopt automation and AI on the defensive side.

🔍 In this Humans of Talos interview, Senior Vulnerability Researcher Philippe Laulheret explains how his lifelong curiosity and Capture The Flag experience led him from French engineering school to a career in ethical hacking. He describes selecting research targets, reverse engineering techniques, and memorable tests—like bypassing a fingerprint reader with a green onion—to find flaws before adversaries exploit them. Philippe also contrasts the methodical reality of research with movie portrayals and outlines his path through industry roles to Talos.

🔷 Google presents a cluster-level reliability framework for TPU superpods that treats thousands of chips as collective units rather than independent instances. The framework replaces instance-level MTBF thinking with a probabilistic, topology-aware model (binomial distribution) to guarantee contiguous healthy cubes for massive training runs. Using Ironwood, Google shows a 95% confidence block of 130/144 cubes—an 8,320-chip domain—while allowing remaining capacity for heterogeneous workloads. Combined with framework resilience and multi-tier checkpointing, this model is engineered to maximize scheduling goodput for hero jobs.

📌 At Cosmos Conf 2026 Microsoft outlined how AI is transforming application and database design, arguing data platforms must become systems of reasoning that handle prompts, memory, and evolving context. Leaders from OpenAI, Vercel, and Walmart stressed the need for serverless instant scalability, integrated caching, low-latency global distribution, and developer cost visibility. Demos and customer stories highlighted patterns like vector search, change feed, and role-based governance to deliver real-world, low-latency AI experiences.

🔍 The Google Threat Intelligence Group (GTIG) reported the first confirmed instance of an AI-crafted zero-day exploit observed in the wild. The researchers identified a Python-based exploit that bypasses two-factor authentication in an open-source web administration tool and disclosed the flaw to the vendor to limit mass exploitation. GTIG found artifacts in the code—help text, a hallucinated CVSS score and textbook LLM-style constructs—consistent with large language model generation, and noted broader AI abuse by threat actors including misuse of Gemini and agentic tooling.