The MRTD Research Desk produces MRTD.NET's coverage of crypto-security incidents, cyber news, and search/growth. We work AI-assisted under human editorial oversight, verify against primary sources and on-chain data, and link everything we cite. See our editorial standards and corrections policy.
llms.txt — a proposed markdown 'map' for language models — now sits on roughly one in ten sites. But in 90 days of 500M+ AI-bot visits, only a few hundred fetched it, and Google has explicitly said it doesn't use it. Here's the honest split: near-zero value for AI search, real value for developer tooling.
A nonprofit's 27-year-old domain was moved into a stranger's account in minutes, with the DNS wiped — even though the account had dual two-factor auth and ownership protection turned on. The transfer didn't break the security; it bypassed it entirely, through GoDaddy's own support desk. That's the threat model everyone forgets.
Reddit appears in roughly 93% of AI-search opportunities and is the single most-cited source for Google AI Overviews and Perplexity. Yet most of that influence is invisible to users, and in 2025 the numbers swung wildly — ChatGPT's Reddit citations fell from ~60% to ~10% in six weeks. Here is what site owners should actually take from it.
Researchers found a dormant facial-recognition feature called 'NameTag' buried in Meta's AI companion app — face-matching, local databases, the works — shipped to an app with 50M+ installs. Meta removed it within about 24 hours of WIRED's report and insisted it was never enabled. Whether you believe that, 'built but switched off' is its own kind of warning.
A DOJ-led public-private operation disabled more than 1.4 million Southeast Asian scam accounts and made 63 arrests — yet froze under $4 million in crypto. The viral '$3 billion frozen' figure is wrong, and the real gap between accounts taken down and money recovered is the actual lesson.
Recent Chrome builds ship Gemini Nano — a ~4GB on-device AI model — downloaded in the background to power new built-in browser APIs. Running locally is a genuine privacy win, but a multi-gigabyte model installed without a clear prompt raises a fair consent question. Here is what is actually on your machine.
The US seized 127,271 bitcoin — about 0.64% of all the bitcoin ever mined — from Cambodia's Prince Group, the largest forfeiture in DOJ history. The coins sat untouched since 2020, yet were fully traced. Now the fight is over where $15 billion goes: a federal reserve, or the scam's victims.
Signal has again refused to comply with a reported UK plan to scan devices for illegal imagery, bundled with age verification. The objection is not about any single law — it is that client-side scanning is fundamentally incompatible with end-to-end encryption.
DeFi project DIP Protocol was drained of roughly $111,000 on BNB Chain through a token-transfer bug that let an attacker double-count pool reserves. Here is the class of flaw involved and why it keeps reappearing.
New vendor analyses spanning tens of thousands of domains suggest page-level SEO strength still correlates with being cited in ChatGPT, Perplexity and AI Overviews, yet a large share of AI recommendations never appear in traditional rank trackers. We separate the load-bearing tactics from the GEO hype.
SEO practitioners report large traffic gains from pruning junk indexed pages to free Google's "crawl budget." We break down what crawl budget actually is per Google's own docs, who it genuinely matters for, and a concrete reclamation playbook — while staying skeptical of the headline +67% figure.
An attacker exploited a settlement-boundary flaw in Aztec Connect's abandoned RollupProcessorV3 contract on June 14, 2026, draining roughly $2.19M in ETH and stablecoins. The privacy bridge was sunset in March 2023 and Aztec Labs holds no keys to pause it. The case is a reminder that "deprecated" is not "safe."