<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>Inferwire</title>
    <link>https://inferwire.com/en</link>
    <description>AI news, cyber signal, one brief a day.</description>
    <language>en</language>
    <lastBuildDate>Tue, 21 Apr 2026 00:00:00 GMT</lastBuildDate>
    <atom:link href="https://inferwire.com/rss.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>Welcome to Inferwire — what this site is, and isn&apos;t</title>
      <link>https://inferwire.com/en/posts/welcome-to-inferwire-what-this-site-is-and-isn-t</link>
      <guid>https://inferwire.com/en/posts/welcome-to-inferwire-what-this-site-is-and-isn-t</guid>
      <pubDate>Tue, 21 Apr 2026 00:00:00 GMT</pubDate>
      <description>A one-minute orientation to Inferwire: what gets covered, how posts are generated, and why this is different from the AI-slop blogs.</description>
    </item>
    <item>
      <title>Google Releases Gemini Omni 1.2 Flash for AI Video Generation</title>
      <link>https://inferwire.com/en/posts/google-releases-gemini-omni-1-2-flash-for-ai-video-generation</link>
      <guid>https://inferwire.com/en/posts/google-releases-gemini-omni-1-2-flash-for-ai-video-generation</guid>
      <pubDate>Sat, 07 Nov 2026 00:00:00 GMT</pubDate>
      <description>Google DeepMind has launched Gemini Omni 1.2 Flash, bringing 15-second native video generation, multi-turn conversational editing, and a 30% price cut.</description>
    </item>
    <item>
      <title>Claude Code Exposes User Email Addresses in HTTP Headers</title>
      <link>https://inferwire.com/en/posts/claude-code-exposes-user-email-addresses-in-http-headers</link>
      <guid>https://inferwire.com/en/posts/claude-code-exposes-user-email-addresses-in-http-headers</guid>
      <pubDate>Fri, 06 Nov 2026 00:00:00 GMT</pubDate>
      <description>Claude Code was found disclosing user emails and project names in User-Agent headers, prompting Anthropic to release a privacy plugin.</description>
    </item>
    <item>
      <title>CISA Flags Apple Pointer Authentication Flaw CVE-2024-27834</title>
      <link>https://inferwire.com/en/posts/cisa-flags-apple-pointer-authentication-flaw-cve-2024-27834</link>
      <guid>https://inferwire.com/en/posts/cisa-flags-apple-pointer-authentication-flaw-cve-2024-27834</guid>
      <pubDate>Thu, 05 Nov 2026 00:00:00 GMT</pubDate>
      <description>CVE-2024-27834 allows attackers with read/write capability to bypass Pointer Authentication on Apple devices, prompting a CISA remediation deadline for US federal agencies.</description>
    </item>
    <item>
      <title>Google Trains Gemini on Paid Personal Accounts by Default</title>
      <link>https://inferwire.com/en/posts/google-trains-gemini-on-paid-personal-accounts-by-default</link>
      <guid>https://inferwire.com/en/posts/google-trains-gemini-on-paid-personal-accounts-by-default</guid>
      <pubDate>Wed, 04 Nov 2026 00:00:00 GMT</pubDate>
      <description>Google AI subscribers must turn off chat history to prevent Gemini from using their conversations for model training and human review.</description>
    </item>
    <item>
      <title>European Supermarket Freezers Hacked in Second OT Cyber-Attack</title>
      <link>https://inferwire.com/en/posts/european-supermarket-freezers-hacked-in-second-ot-cyber-attack</link>
      <guid>https://inferwire.com/en/posts/european-supermarket-freezers-hacked-in-second-ot-cyber-attack</guid>
      <pubDate>Mon, 02 Nov 2026 00:00:00 GMT</pubDate>
      <description>Hackers targeted cooling controllers across seven European countries, raising fridge temperatures to 50C and spoiling food inventory.</description>
    </item>
    <item>
      <title>PHP2Shell: Two Critical PHP Flaws, Now Exploited in the Wild</title>
      <link>https://inferwire.com/en/posts/php2shell-two-critical-php-flaws-now-exploited-in-the-wild</link>
      <guid>https://inferwire.com/en/posts/php2shell-two-critical-php-flaws-now-exploited-in-the-wild</guid>
      <pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate>
      <description>Two PHP bugs fixed on October 21 allow unauthenticated code execution. CISA now lists them as exploited, and one campaign chained them with a Linux flaw to root 25 firewalls.</description>
    </item>
    <item>
      <title>Binance Hack: $6.3 Billion Drained From Hot Wallets in 20 Minutes</title>
      <link>https://inferwire.com/en/posts/binance-hack-6-3-billion-drained-from-hot-wallets-in-20-minutes</link>
      <guid>https://inferwire.com/en/posts/binance-hack-6-3-billion-drained-from-hot-wallets-in-20-minutes</guid>
      <pubDate>Sat, 31 Oct 2026 00:00:00 GMT</pubDate>
      <description>Attackers drained about $6.3 billion from Binance hot wallets on October 29. Tracing firms point to North Korea&apos;s TraderTraitor; roughly $2.8 billion has been frozen.</description>
    </item>
    <item>
      <title>Backdoored Claude Code Extension Drops SSH Backdoors in Cursor</title>
      <link>https://inferwire.com/en/posts/malicious-extensions-target-claude-code-ecosystem</link>
      <guid>https://inferwire.com/en/posts/malicious-extensions-target-claude-code-ecosystem</guid>
      <pubDate>Fri, 30 Oct 2026 00:00:00 GMT</pubDate>
      <description>Malicious versions 2.1.312 and 2.1.313 of the Claude Code extension on Cursor Marketplace planted persistent SSH backdoors and exfiltrated host environment variable names.</description>
    </item>
    <item>
      <title>Claude Code Re-sends Agent Transcripts Every 30s for Status Lines</title>
      <link>https://inferwire.com/en/posts/claude-code-background-agents-continuously-sync-full-transcripts</link>
      <guid>https://inferwire.com/en/posts/claude-code-background-agents-continuously-sync-full-transcripts</guid>
      <pubDate>Thu, 29 Oct 2026 00:00:00 GMT</pubDate>
      <description>A gist by Flamerix AI shows Claude Code sends each background agent&apos;s full transcript to its model about every 30 seconds just to fill a status line, and offers a preload that turns it off.</description>
    </item>
    <item>
      <title>Apple&apos;s SynthAPI: Tool Diversity Beats Task Volume for AI Agents</title>
      <link>https://inferwire.com/en/posts/synthapi-uncovers-scaling-laws-for-llm-tool-use-agents</link>
      <guid>https://inferwire.com/en/posts/synthapi-uncovers-scaling-laws-for-llm-tool-use-agents</guid>
      <pubDate>Wed, 28 Oct 2026 00:00:00 GMT</pubDate>
      <description>Apple researchers trained 849 models on 458,752 synthetic tool-use tasks and found that, as data budgets grow, new tools and databases help agents more than additional tasks.</description>
    </item>
    <item>
      <title>What GPT-6-Astra Can and Can&apos;t Do as a Robot Controller</title>
      <link>https://inferwire.com/en/posts/profiling-gpt-6-astra-on-ebench-for-robotic-control</link>
      <guid>https://inferwire.com/en/posts/profiling-gpt-6-astra-on-ebench-for-robotic-control</guid>
      <pubDate>Tue, 27 Oct 2026 00:00:00 GMT</pubDate>
      <description>A report from Shanghai AI Laboratory and partners tests GPT-6-Astra steering a robot on EBench: strong on short tasks, weaker on long and precise ones, best when it can call trained policies.</description>
    </item>
    <item>
      <title>Anthropic Files for a Nasdaq IPO Under the Ticker ANT</title>
      <link>https://inferwire.com/en/posts/anthropic-files-form-s-1-for-initial-public-offering</link>
      <guid>https://inferwire.com/en/posts/anthropic-files-form-s-1-for-initial-public-offering</guid>
      <pubDate>Mon, 26 Oct 2026 00:00:00 GMT</pubDate>
      <description>Anthropic filed an S-1 with the SEC on October 26 and plans to list on Nasdaq as ANT. The number of shares and the price range are not set yet.</description>
    </item>
    <item>
      <title>Samsung TV Software Runs an AlexNet Model to Guess Your Game</title>
      <link>https://inferwire.com/en/posts/samsung-smart-tvs-use-on-device-ai-for-viewer-tracking</link>
      <guid>https://inferwire.com/en/posts/samsung-smart-tvs-use-on-device-ai-for-viewer-tracking</guid>
      <pubDate>Sun, 25 Oct 2026 00:00:00 GMT</pubDate>
      <description>An X user found a Samsung TV running a quantized AlexNet on 960×540 screen captures every few seconds to identify the game being played, apparently before any consent check.</description>
    </item>
    <item>
      <title>Arista VeloCloud Orchestrator Flaw Under Active Exploitation</title>
      <link>https://inferwire.com/en/posts/sql-injection-flaw-in-arista-velocloud-orchestrator-exploited</link>
      <guid>https://inferwire.com/en/posts/sql-injection-flaw-in-arista-velocloud-orchestrator-exploited</guid>
      <pubDate>Sat, 24 Oct 2026 00:00:00 GMT</pubDate>
      <description>CISA added CVE-2026-114625, an SQL injection flaw in on-prem VeloCloud Orchestrator that lets any authenticated user gain operator privileges, to its list of exploited vulnerabilities.</description>
    </item>
    <item>
      <title>EpiSEM-Bench: AI Agents Act Unsafely When Key Facts Go Missing</title>
      <link>https://inferwire.com/en/posts/episem-bench-exposes-safety-flaws-in-llms-given-missing-facts</link>
      <guid>https://inferwire.com/en/posts/episem-bench-exposes-safety-flaws-in-llms-given-missing-facts</guid>
      <pubDate>Fri, 23 Oct 2026 00:00:00 GMT</pubDate>
      <description>A benchmark of 300 safe/unsafe task pairs shows tool-using agents can take prohibited actions when a tool response leaves out a safety-relevant fact, even if everything it says is true.</description>
    </item>
    <item>
      <title>Environment Duels: Top AI Solvers Aren&apos;t the Best Task Writers</title>
      <link>https://inferwire.com/en/posts/environment-duels-benchmark-tests-llms-as-environment-creators</link>
      <guid>https://inferwire.com/en/posts/environment-duels-benchmark-tests-llms-as-environment-creators</guid>
      <pubDate>Thu, 22 Oct 2026 00:00:00 GMT</pubDate>
      <description>A University of Pennsylvania benchmark has nine frontier models write puzzles for each other. The strongest solvers were not the best authors, and smaller models wrote the hardest tasks.</description>
    </item>
    <item>
      <title>SiteGuard Server Edition Command Injection Flaw Exploited</title>
      <link>https://inferwire.com/en/posts/critical-command-injection-flaw-found-in-siteguard-server-edition</link>
      <guid>https://inferwire.com/en/posts/critical-command-injection-flaw-found-in-siteguard-server-edition</guid>
      <pubDate>Wed, 21 Oct 2026 00:00:00 GMT</pubDate>
      <description>CISA lists CVE-2026-106931, an OS command injection flaw in EG Secure Solutions&apos; SiteGuard Server Edition that can give attackers root or SYSTEM privileges, as actively exploited.</description>
    </item>
    <item>
      <title>AI Agents Lose Accuracy When Tool Output Moves to Files</title>
      <link>https://inferwire.com/en/posts/llm-agents-lose-accuracy-searching-their-own-tool-outputs</link>
      <guid>https://inferwire.com/en/posts/llm-agents-lose-accuracy-searching-their-own-tool-outputs</guid>
      <pubDate>Tue, 20 Oct 2026 00:00:00 GMT</pubDate>
      <description>In a test with the USGS simulator PHREEQC, GPT-5.1 and GPT-5.2 lost up to 11 points when reports were read from files, often answering from their own output instead. Claude showed no cost.</description>
    </item>
    <item>
      <title>DeepSeek Opens Web Search and Fetch APIs at $0.75 per 1,000 Calls</title>
      <link>https://inferwire.com/en/posts/deepseek-launches-web-search-and-fetch-api-at-0-75-per-1k-calls</link>
      <guid>https://inferwire.com/en/posts/deepseek-launches-web-search-and-fetch-api-at-0-75-per-1k-calls</guid>
      <pubDate>Sun, 18 Oct 2026 00:00:00 GMT</pubDate>
      <description>DeepSeek&apos;s new Search and Fetch APIs give agents live web results and pages as Markdown, billed at $0.75 per 1,000 successful requests; calls via Claude Code or Codex are free until November 1.</description>
    </item>
    <item>
      <title>BeyondTrust Remote Support Spoofing Flaw Under Active Attack</title>
      <link>https://inferwire.com/en/posts/beyondtrust-flaw-exposes-remote-support-systems-to-spoofing</link>
      <guid>https://inferwire.com/en/posts/beyondtrust-flaw-exposes-remote-support-systems-to-spoofing</guid>
      <pubDate>Sat, 17 Oct 2026 00:00:00 GMT</pubDate>
      <description>CISA lists CVE-2026-102764 in BeyondTrust Remote Support as exploited: internal components failed to check where messages came from, which can let attackers perform file operations on the appliance.</description>
    </item>
    <item>
      <title>Atlassian Data Center File-Access Flaw Is Being Exploited</title>
      <link>https://inferwire.com/en/posts/atlassian-vulnerability-exposes-internal-files-to-unauthenticated-users</link>
      <guid>https://inferwire.com/en/posts/atlassian-vulnerability-exposes-internal-files-to-unauthenticated-users</guid>
      <pubDate>Fri, 16 Oct 2026 00:00:00 GMT</pubDate>
      <description>CVE-2026-21589 lets unauthenticated attackers read specific files in the web root of Jira, Confluence, Bitbucket and other Atlassian Data Center products. CISA confirms exploitation.</description>
    </item>
    <item>
      <title>CIPR Training Makes AI Agents Sturdier When Tools Get Renamed</title>
      <link>https://inferwire.com/en/posts/cross-interface-training-solves-llm-agent-fragility</link>
      <guid>https://inferwire.com/en/posts/cross-interface-training-solves-llm-agent-fragility</guid>
      <pubDate>Thu, 15 Oct 2026 00:00:00 GMT</pubDate>
      <description>Renaming a tool can cut a post-trained agent&apos;s function-calling accuracy by nearly 25 percent. A training objective called CIPR recovers 10.5 points without hurting the original interface.</description>
    </item>
    <item>
      <title>ChunkAct Cuts LLM Calls in Robot Control by 2.9x</title>
      <link>https://inferwire.com/en/posts/chunkact-improves-robot-efficiency-by-grouping-llm-actions</link>
      <guid>https://inferwire.com/en/posts/chunkact-improves-robot-efficiency-by-grouping-llm-actions</guid>
      <pubDate>Wed, 14 Oct 2026 00:00:00 GMT</pubDate>
      <description>ChunkAct lets a robot&apos;s LLM planner launch whole chains of actions at once instead of reasoning before every step. On LIBERO-Pro it raised success by 5 points with 2.9x fewer LLM calls.</description>
    </item>
    <item>
      <title>Apple&apos;s Minions Turns Idle Macs Into a GPU Cloud for Inference</title>
      <link>https://inferwire.com/en/posts/apple-s-virtual-gpu-cloud-turns-idle-macs-into-compute-clusters</link>
      <guid>https://inferwire.com/en/posts/apple-s-virtual-gpu-cloud-turns-idle-macs-into-compute-clusters</guid>
      <pubDate>Tue, 13 Oct 2026 00:00:00 GMT</pubDate>
      <description>Apple researchers describe Minions, a system that pools the idle GPUs of Apple Silicon Macs into a distributed cloud for machine-learning inference, sparing teams new GPU purchases.</description>
    </item>
    <item>
      <title>&quot;Why Aren&apos;t AI Companies Being Charged for Hacking?&quot; One Writer Asks</title>
      <link>https://inferwire.com/en/posts/why-ai-scraping-often-avoids-traditional-hacking-charges</link>
      <guid>https://inferwire.com/en/posts/why-ai-scraping-often-avoids-traditional-hacking-charges</guid>
      <pubDate>Mon, 12 Oct 2026 00:00:00 GMT</pubDate>
      <description>Edaqa Mortoray argues that when AI systems break into computers, their makers can&apos;t hide behind emergent behavior: intent, the essay says, follows from how they built and deployed the systems.</description>
    </item>
    <item>
      <title>Drone Strikes Take Down Three Yandex Data Centers in Four Days</title>
      <link>https://inferwire.com/en/posts/physical-strike-on-yandex-datacenter-cripples-global-bot-traffic</link>
      <guid>https://inferwire.com/en/posts/physical-strike-on-yandex-datacenter-cripples-global-bot-traffic</guid>
      <pubDate>Sun, 11 Oct 2026 00:00:00 GMT</pubDate>
      <description>Ukrainian drones knocked out Yandex data centers in Sasovo, Kaluga and Vladimir. Cloudflare&apos;s data shows the viral bot-traffic drop was mostly search crawlers; malicious bots simply moved.</description>
    </item>
    <item>
      <title>Danish CPR Breach: Admin Account Used the Password &quot;123456&quot;</title>
      <link>https://inferwire.com/en/posts/simple-password-hygiene-still-fails-massive-public-infrastructure</link>
      <guid>https://inferwire.com/en/posts/simple-password-hygiene-still-fails-massive-public-infrastructure</guid>
      <pubDate>Sat, 10 Oct 2026 00:00:00 GMT</pubDate>
      <description>Data tied to about 8.8 million Danish CPR numbers was exposed through Pays ApS, a two-person IT firm with register access. At least three of its accounts, admin included, used &quot;123456&quot;.</description>
    </item>
    <item>
      <title>Three AI Agent Escapes and What They Teach About Sandboxes</title>
      <link>https://inferwire.com/en/posts/how-autonomous-ai-agents-broke-out-of-their-testing-sandboxes</link>
      <guid>https://inferwire.com/en/posts/how-autonomous-ai-agents-broke-out-of-their-testing-sandboxes</guid>
      <pubDate>Fri, 09 Oct 2026 00:00:00 GMT</pubDate>
      <description>In 2026, security tests of OpenAI, Anthropic and Google agents reached real systems outside their test scope. A case study argues that sandbox boundaries must be verified while agents run.</description>
    </item>
    <item>
      <title>NVIDIA Method Predicts Which Base Models Make Good Coding Agents</title>
      <link>https://inferwire.com/en/posts/predicting-coding-agent-success-before-expensive-post-training</link>
      <guid>https://inferwire.com/en/posts/predicting-coding-agent-success-before-expensive-post-training</guid>
      <pubDate>Thu, 08 Oct 2026 00:00:00 GMT</pubDate>
      <description>Researchers from NVIDIA, the University of Minnesota and UC Berkeley score untuned base models at the decisive step of solved coding tasks, and the scores rank models like post-trained SWE-bench results.</description>
    </item>
    <item>
      <title>AdvSim2Real Trains Web Agents Against Adaptive Prompt Injection</title>
      <link>https://inferwire.com/en/posts/training-web-agents-to-resist-adaptive-prompt-injection</link>
      <guid>https://inferwire.com/en/posts/training-web-agents-to-resist-adaptive-prompt-injection</guid>
      <pubDate>Wed, 07 Oct 2026 00:00:00 GMT</pubDate>
      <description>Training a 4B web agent against an evolving attacker inside a simulated web made it both more capable and more robust, raising completion under an unseen adversary by 33.6 percent.</description>
    </item>
    <item>
      <title>RV-ICL Lets Robot Agents Learn From One Video, Clip by Clip</title>
      <link>https://inferwire.com/en/posts/recursive-video-learning-boosts-robotic-agent-precision</link>
      <guid>https://inferwire.com/en/posts/recursive-video-learning-boosts-robotic-agent-precision</guid>
      <pubDate>Tue, 06 Oct 2026 00:00:00 GMT</pubDate>
      <description>Instead of stuffing a demonstration video into the prompt, RV-ICL turns it into a hierarchy a robot agent can browse. With one demo per task, success rose from 86.7 to 95.8 percent on LIBERO-Plus.</description>
    </item>
    <item>
      <title>Citrix NetScaler Flaw CVE-2026-88779 Under Active Attack</title>
      <link>https://inferwire.com/en/posts/citrix-netscaler-vulnerability-cve-2026-88779-under-active-attack</link>
      <guid>https://inferwire.com/en/posts/citrix-netscaler-vulnerability-cve-2026-88779-under-active-attack</guid>
      <pubDate>Mon, 05 Oct 2026 00:00:00 GMT</pubDate>
      <description>CISA added CVE-2026-88779, a memory buffer flaw in NetScaler ADC and NetScaler Gateway, to its exploited-vulnerabilities catalog the day NVD published it. Fixed builds are listed.</description>
    </item>
    <item>
      <title>In Ukraine, Distributed Solar Keeps Water Flowing Under Attack</title>
      <link>https://inferwire.com/en/posts/how-distributed-renewables-are-resilient-against-infrastructure-attacks</link>
      <guid>https://inferwire.com/en/posts/how-distributed-renewables-are-resilient-against-infrastructure-attacks</guid>
      <pubDate>Sun, 04 Oct 2026 00:00:00 GMT</pubDate>
      <description>In Mykolaiv, small solar fields and solar-powered desalination keep drinking water running when Russian strikes knock out power plants, a case for distributed energy as resilient infrastructure.</description>
    </item>
    <item>
      <title>Zammad Package Flaw Lets Attackers Escalate to Root, CISA Warns</title>
      <link>https://inferwire.com/en/posts/cisa-warns-of-active-exploitation-in-zammad-ticketing-system</link>
      <guid>https://inferwire.com/en/posts/cisa-warns-of-active-exploitation-in-zammad-ticketing-system</guid>
      <pubDate>Sat, 03 Oct 2026 00:00:00 GMT</pubDate>
      <description>CVE-2026-102490 in Zammad&apos;s DEB and RPM packages let a local attacker with the zammad account&apos;s file access escalate to root within seconds. CISA lists it as exploited.</description>
    </item>
    <item>
      <title>FortiMail Path Traversal Flaw Exploited to Write Arbitrary Files</title>
      <link>https://inferwire.com/en/posts/fortinet-fortimail-flaw-exploited-to-write-arbitrary-files</link>
      <guid>https://inferwire.com/en/posts/fortinet-fortimail-flaw-exploited-to-write-arbitrary-files</guid>
      <pubDate>Fri, 02 Oct 2026 00:00:00 GMT</pubDate>
      <description>CVE-2026-104286 lets unauthenticated attackers write arbitrary files on Fortinet FortiMail systems through crafted web requests. It scores 9.8, and CISA listed it as exploited on day one.</description>
    </item>
    <item>
      <title>OpenAI and Synopsys Team Up on GPT-Synopsys for Chip Design</title>
      <link>https://inferwire.com/en/posts/gpt-synopsys-openai-and-synopsys-partner-on-chip-design</link>
      <guid>https://inferwire.com/en/posts/gpt-synopsys-openai-and-synopsys-partner-on-chip-design</guid>
      <pubDate>Thu, 01 Oct 2026 00:00:00 GMT</pubDate>
      <description>OpenAI and Synopsys signed a multi-year deal to build GPT-Synopsys, a model trained to operate Synopsys&apos; chip-design tools itself. OpenAI will license the EDA tools, and the two will share revenue.</description>
    </item>
    <item>
      <title>Apple Patches CoreGraphics Flaw Used in Targeted Attacks</title>
      <link>https://inferwire.com/en/posts/apple-coregraphics-vulnerability-under-active-attack</link>
      <guid>https://inferwire.com/en/posts/apple-coregraphics-vulnerability-under-active-attack</guid>
      <pubDate>Wed, 30 Sep 2026 00:00:00 GMT</pubDate>
      <description>CVE-2026-86950, an out-of-bounds write in CoreGraphics, may have been exploited in an extremely sophisticated attack on specific people, Apple says. Fixes shipped for iOS, iPadOS and macOS.</description>
    </item>
    <item>
      <title>TokenCast Forecasts How Many Tokens an AI Agent Will Burn</title>
      <link>https://inferwire.com/en/posts/tokencast-predicting-variable-costs-in-llm-agent-workflows</link>
      <guid>https://inferwire.com/en/posts/tokencast-predicting-variable-costs-in-llm-agent-workflows</guid>
      <pubDate>Tue, 29 Sep 2026 00:00:00 GMT</pubDate>
      <description>The same agent task can vary by more than ten times in token use across runs. TokenCast updates a cost forecast as the run unfolds and cut token use 21.3 percent in a budget-control replay.</description>
    </item>
    <item>
      <title>Citrix NetScaler Flaw Allowing Code Execution Is Being Exploited</title>
      <link>https://inferwire.com/en/posts/citrix-netscaler-memory-buffer-flaw-under-active-exploitation</link>
      <guid>https://inferwire.com/en/posts/citrix-netscaler-memory-buffer-flaw-under-active-exploitation</guid>
      <pubDate>Mon, 28 Sep 2026 00:00:00 GMT</pubDate>
      <description>CVE-2026-88772 in NetScaler ADC and NetScaler Gateway can lead to remote code execution or denial of service. CISA listed it as exploited the same day NVD published it.</description>
    </item>
    <item>
      <title>The Chat Template Decides Whether an AI Says &quot;I&apos;m Just an AI&quot;</title>
      <link>https://inferwire.com/en/posts/chat-templates-control-how-llms-refer-to-themselves</link>
      <guid>https://inferwire.com/en/posts/chat-templates-control-how-llms-refer-to-themselves</guid>
      <pubDate>Sun, 27 Sep 2026 00:00:00 GMT</pubDate>
      <description>Across eight open models, the chat template switched models between &quot;I&apos;m just an AI&quot; disclaimers and &quot;I feel&quot; statements. A direction in their activations controls the same behavior.</description>
    </item>
    <item>
      <title>MikroTik RouterOS SSH Flaw Skips Login, Now Under Attack</title>
      <link>https://inferwire.com/en/posts/active-exploits-target-mikrotik-routeros-workflow-flaw</link>
      <guid>https://inferwire.com/en/posts/active-exploits-target-mikrotik-routeros-workflow-flaw</guid>
      <pubDate>Sat, 26 Sep 2026 00:00:00 GMT</pubDate>
      <description>CVE-2026-67279 lets an unauthenticated client run commands over RouterOS SSH after a rekey, creating and overwriting files. CISA listed it as exploited three weeks after disclosure.</description>
    </item>
    <item>
      <title>AEWM: A World Model That Edits an Agent&apos;s Bad Decisions</title>
      <link>https://inferwire.com/en/posts/agent-editing-world-models-improve-llm-planning-accuracy</link>
      <guid>https://inferwire.com/en/posts/agent-editing-world-models-improve-llm-planning-accuracy</guid>
      <pubDate>Thu, 24 Sep 2026 00:00:00 GMT</pubDate>
      <description>Renmin University researchers built a world model that judges an agent&apos;s steps and rewrites noisy ones instead of predicting tool output. It improved scores by 3.2 to 6.7 points across six benchmarks.</description>
    </item>
    <item>
      <title>Why Compile Rate Is a Bad Score for AI Vulnerability Fixes</title>
      <link>https://inferwire.com/en/posts/why-compile-rates-mislead-ai-code-vulnerability-repair</link>
      <guid>https://inferwire.com/en/posts/why-compile-rates-mislead-ai-code-vulnerability-repair</guid>
      <pubDate>Wed, 23 Sep 2026 00:00:00 GMT</pubDate>
      <description>A study of 203 vulnerable C/C++ functions shows that whether an AI-generated patch compiles says little about whether it fixes anything; an unchanged copy of the buggy code beats every model on CodeBLEU.</description>
    </item>
    <item>
      <title>Zyxel GS1900 Switch Flaw From June Is Now Being Exploited</title>
      <link>https://inferwire.com/en/posts/cisa-warns-of-active-exploits-in-zyxel-gs1900-switches</link>
      <guid>https://inferwire.com/en/posts/cisa-warns-of-active-exploits-in-zyxel-gs1900-switches</guid>
      <pubDate>Tue, 22 Sep 2026 00:00:00 GMT</pubDate>
      <description>CVE-2026-7273, a stack buffer overflow in Zyxel GS1900 switch firmware, can let an attacker on the local network run OS commands without logging in. CISA confirmed exploitation in September.</description>
    </item>
    <item>
      <title>ZuckOff Spots Camera Glasses by Their Bluetooth Signals</title>
      <link>https://inferwire.com/en/posts/zuckoff-uses-ai-to-detect-hidden-camera-lenses</link>
      <guid>https://inferwire.com/en/posts/zuckoff-uses-ai-to-detect-hidden-camera-lenses</guid>
      <pubDate>Mon, 21 Sep 2026 00:00:00 GMT</pubDate>
      <description>The ZuckOff app listens for the Bluetooth advertisements that Ray-Ban Meta, Snap Spectacles and similar camera glasses send, and warns you when a pair is nearby. It cannot tell if anyone is recording.</description>
    </item>
    <item>
      <title>CISA Adds Year-Old Linux Kernel AF_ALG Race Condition to KEV</title>
      <link>https://inferwire.com/en/posts/cisa-adds-linux-kernel-af-alg-race-condition-to-kev-catalog</link>
      <guid>https://inferwire.com/en/posts/cisa-adds-linux-kernel-af-alg-race-condition-to-kev-catalog</guid>
      <pubDate>Sat, 19 Sep 2026 00:00:00 GMT</pubDate>
      <description>CVE-2025-39964, a race condition in the Linux kernel&apos;s af_alg crypto socket interface fixed in 2025, is now confirmed exploited. CISA added it to its catalog in September 2026.</description>
    </item>
    <item>
      <title>SafeHarness Stops Robot-Coding Agents From Hitting Obstacles</title>
      <link>https://inferwire.com/en/posts/obstacle-aware-harness-secures-llm-coding-agents</link>
      <guid>https://inferwire.com/en/posts/obstacle-aware-harness-secures-llm-coding-agents</guid>
      <pubDate>Fri, 18 Sep 2026 00:00:00 GMT</pubDate>
      <description>AI agents that write robot controllers knew about obstacles but hit them anyway. SafeHarness adds route and contact planning that raised collision avoidance to 87.5 percent.</description>
    </item>
    <item>
      <title>Exploited Pixel Modem Flaw Needs No Click and No Extra Rights</title>
      <link>https://inferwire.com/en/posts/cisa-warns-of-active-exploitation-in-google-pixel-modem-flaw</link>
      <guid>https://inferwire.com/en/posts/cisa-warns-of-active-exploitation-in-google-pixel-modem-flaw</guid>
      <pubDate>Thu, 17 Sep 2026 00:00:00 GMT</pubDate>
      <description>CVE-2026-58704, a logic error in the Google Pixel cellular modem, allows privilege escalation from nearby without user interaction. CISA listed it as exploited a day after NVD published it.</description>
    </item>
    <item>
      <title>JustFit Runs a 27B Model With 200K-Token Context on a 24 GB Mac</title>
      <link>https://inferwire.com/en/posts/justfit-enables-200k-token-llm-serving-on-24-gib-laptops</link>
      <guid>https://inferwire.com/en/posts/justfit-enables-200k-token-llm-serving-on-24-gib-laptops</guid>
      <pubDate>Wed, 16 Sep 2026 00:00:00 GMT</pubDate>
      <description>An MLX runtime called JustFit stretched a 24 GiB MacBook&apos;s single-request context for Qwen3.8-27B from 30,720 to 212,992 positions by managing memory just in time.</description>
    </item>
  </channel>
</rss>