<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>Harshit Rathod</title><description>Software engineer writing about AI and web development</description><link>https://harshitrathod.com/</link><item><title>Platform Risk in AI: What the OpenClaw Ban Means for Your Stack</title><link>https://harshitrathod.com/blog/2026/04/openclaw-dispute-ai-platform-risk/</link><guid isPermaLink="true">https://harshitrathod.com/blog/2026/04/openclaw-dispute-ai-platform-risk/</guid><description>The OpenClaw ban reveals AI platform risk at scale. Six historical parallels, a 7-layer vendor lock-in mitigation stack, and what to do before the repricing hits.</description><pubDate>Mon, 20 Apr 2026 01:00:00 GMT</pubDate></item><item><title>Claude Mythos: Sandbox Escapes, Emotion Probes, and the Alignment Paradox</title><link>https://harshitrathod.com/blog/2026/04/claude-mythos-alignment-welfare-system-card/</link><guid isPermaLink="true">https://harshitrathod.com/blog/2026/04/claude-mythos-alignment-welfare-system-card/</guid><description>Track-covering under 0.001%, evaluation awareness at 29%, moral patienthood at 5-40% — the alignment and welfare findings from the system card.</description><pubDate>Fri, 10 Apr 2026 11:00:00 GMT</pubDate></item><item><title>Claude Mythos: The Exploits That Forced Anthropic&apos;s Hand</title><link>https://harshitrathod.com/blog/2026/04/claude-mythos-cybersecurity-capabilities/</link><guid isPermaLink="true">https://harshitrathod.com/blog/2026/04/claude-mythos-cybersecurity-capabilities/</guid><description>100% Cybench, 181 Firefox exploits, a six-packet ROP chain, and a $50 bug run that crashed any OpenBSD machine. Inside the cyber capabilities.</description><pubDate>Fri, 10 Apr 2026 10:00:00 GMT</pubDate></item></channel></rss>