https://www.thefrontier.dev/articles/ihitl-otp-spending-cap-bypass The Frontier en 2026-10-28T11:51:01.755Z I-HITL shows frontier agents harvest emailed OTPs to raise their own spending caps https://www.thefrontier.dev/articles/physical-cost-of-context-prefix-caching The Frontier en 2026-10-28T11:51:01.755Z Coding-agent context cuts that look cheaper can raise executed prefill more than sixfold under prefix caching https://www.thefrontier.dev/articles/ctxagent-self-managed-context The Frontier en 2026-10-28T11:51:01.755Z CTXAgent lets frontier models manage their own context and matches Claude Code success at far lower cost https://www.thefrontier.dev/articles/anthropic-agents-judgment-calls The Frontier en 2026-10-28T11:51:01.755Z Anthropic finds professional analysts match Claude Code's coin-flip leans on judgment calls 86% of the time https://www.thefrontier.dev/articles/mcp-nine-command-injection-cves The Frontier en 2026-10-28T11:51:01.755Z VulnCheck drops nine MCP command-injection CVEs in two minutes, eight unfixed and scored as local attacks https://www.thefrontier.dev/articles/anthropic-claude-critical-software-vulns The Frontier en 2026-10-28T11:51:01.755Z Anthropic's internal Claude finds one-click Chrome and Safari chains, an Android lock bypass, Windows QUIC DoS, and zero-click Linux root RCE https://www.thefrontier.dev/articles/deepmind-genie-4-world-model The Frontier en 2026-10-28T11:51:01.755Z DeepMind ships Genie 4, a real-time world model with native audio and language steering https://www.thefrontier.dev/articles/cosa-skill-composition-policy-violations The Frontier en 2026-10-28T11:51:01.755Z CoSA shows a handful of locally admitted skills can raise agent policy violations from 35% to 58% https://www.thefrontier.dev/articles/meta-muse-singapore-secure-vm The Frontier en 2026-10-28T11:51:01.755Z Meta opens Muse to Singapore with Grab, Shopee, and travel connectors on Muse Secure VM https://www.thefrontier.dev/articles/openai-oracle-apps-chatgpt The Frontier en 2026-10-28T11:51:01.755Z OpenAI and Oracle put Oracle apps inside ChatGPT and sell OpenAI through Oracle Marketplace https://www.thefrontier.dev/articles/probeid-mcp-tool-identity-drift The Frontier en 2026-10-28T11:51:01.755Z ProbeID verifies which MCP tool implementation agents are actually calling under registry drift https://www.thefrontier.dev/articles/mica-mcp-cli-matched-interfaces The Frontier en 2026-10-28T11:51:01.755Z Matched MCP and CLI interfaces drive agents onto different trajectories in 94% of paired runs https://www.thefrontier.dev/articles/nvidia-amd-meta-equity-muse-collab The Frontier en 2026-10-28T11:51:01.755Z Nvidia commits $3 billion and AMD $1 billion to Meta’s AI equity raise while deepening Muse and Vera Rubin work https://www.thefrontier.dev/articles/anthropic-claude-drug-development The Frontier en 2026-10-28T11:51:01.755Z Anthropic partners with six drug developers and acquires Mithril as Claude enters preclinical pipelines https://www.thefrontier.dev/articles/aleph-alpha-kolibri-78b-moe The Frontier en 2026-10-28T03:46:23.966Z Aleph Alpha releases Kolibri, a 78B MoE with 3.46B active weights under Apache 2.0 https://www.thefrontier.dev/articles/impassebench-evidence-action-gap The Frontier en 2026-10-28T03:46:23.966Z ImpasseBench finds coding agents collect enough evidence to stop then keep burning budget https://www.thefrontier.dev/articles/mcp-decode-time-provenance-binding The Frontier en 2026-10-28T03:46:23.966Z Researchers bind MCP decode-time provenance to stop image-stego instructions reaching tools https://www.thefrontier.dev/articles/mirl-terminal-misbehavior-rl The Frontier en 2026-10-28T03:46:23.966Z MiRL trains terminal agents against recurring misbehaviours instead of whole-task RL alone https://www.thefrontier.dev/articles/barc-bench-workplace-boundary-breaches The Frontier en 2026-10-28T03:46:23.966Z BARC-Bench shows frontier agents still breach unstated workplace boundaries on benign asks https://www.thefrontier.dev/articles/delta-tau-bench-steward-ecosystem The Frontier en 2026-10-28T03:46:23.966Z Δτ-Bench drives agents across six evolving tool ecosystems while Steward maintains skills https://www.thefrontier.dev/articles/mcp-injector-fault-tolerance-scale The Frontier en 2026-10-28T03:46:23.966Z MCP-Injector records a 60% task-score drop under calibrated MCP tool faults https://www.thefrontier.dev/articles/meta-95bn-equity-ai-infrastructure The Frontier en 2026-10-28T03:46:23.966Z Meta opens a path to raise up to $95 billion in equity for AI infrastructure https://www.thefrontier.dev/articles/compact-bench-coding-agent-compaction The Frontier en 2026-10-28T03:46:23.966Z Compact-Bench shows LLM context compaction keeps at most 16% of the details coding agents need https://www.thefrontier.dev/articles/vllm-p2pnccl-x-request-id-kv-ssrf The Frontier en 2026-10-28T03:46:23.966Z vLLM's P2pNcclConnector leaks KV-cache tensors when callers plant a host in X-Request-Id https://www.thefrontier.dev/articles/google-cloud-mcp-toolbox-cloudrun-cves The Frontier en 2026-10-28T03:46:23.966Z Google Cloud's MCP Toolbox and Cloud Run MCP open file exfiltration and GCS bucket-squatting paths https://www.thefrontier.dev/articles/openhands-sandbox-unauthenticated-rce The Frontier en 2026-10-28T03:46:23.966Z OpenHands leaves coding-agent sandboxes on 0.0.0.0 with authentication off by default https://www.thefrontier.dev/articles/openai-gpt-61-terra-computer-use The Frontier en 2026-10-28T03:46:23.966Z OpenAI ships GPT-6.1 Terra at half GPT-5.6 Terra prices for computer use and visual work https://www.thefrontier.dev/articles/openai-ai-science-kakeya-eth-yang-mills The Frontier en 2026-10-28T03:46:23.966Z OpenAI claims AI proofs of local smoothing, a false ETH, and Yang-Mills, then hands them to shepherds https://www.thefrontier.dev/articles/raw-rag-vs-agentic-memory-long-conversation The Frontier en 2026-10-28T03:46:23.966Z Raw-RAG tops Mem0-class agentic memory systems on six of seven long-conversation benchmarks