{"id":1026,"date":"2026-11-07T07:15:39","date_gmt":"2026-11-07T07:15:39","guid":{"rendered":"https:\/\/rimsift.com\/?page_id=1026"},"modified":"2026-11-07T07:15:39","modified_gmt":"2026-11-07T07:15:39","slug":"why-rimsift","status":"publish","type":"page","link":"https:\/\/rimsift.com\/index.php\/why-rimsift\/","title":{"rendered":"Why RimSift"},"content":{"rendered":"<p><em>Draft copy imported from the RimSift designer handover. Codes such as G4 refer to the graphics list. Text in [square brackets] is a placeholder or an open decision. Replace this block with the designed layout.<\/em><\/p>\n<ul>\n<li>Eyebrow: WHY RIMSIFT \u00b7 Headline: Approved is not the same as safe.<\/li>\n<li>Sub: Most security tools decide whether an AI agent may connect, then stop watching. These are the eight gaps that leaves, and how RimSift closes each one.<\/li>\n<li>One full-width row per gap, alternating sides: number, title, two-line explanation, and its G3 diagram. Each diagram has two rows: <strong>Today<\/strong> (ends in a red outcome box) and <strong>With RimSift<\/strong> (amber rim labelled with the check, ends green or Blocked).<\/li>\n<\/ul>\n<table>\n<thead>\n<tr>\n<th>#<\/th>\n<th>Title<\/th>\n<th>Explanation<\/th>\n<th>Today row<\/th>\n<th>With RimSift row (check label)<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>G3.1<\/td>\n<td>Nobody watching the agent<\/td>\n<td>Agents are approved once, then act with no one checking each call.<\/td>\n<td>Agent \u2192 approved tool \u2192 unchecked action<\/td>\n<td>Agent \u2192 rim (GOVERNANCE) \u2192 tool, each call checked<\/td>\n<\/tr>\n<tr>\n<td>G3.2<\/td>\n<td>Tools that change after approval<\/td>\n<td>A tool&#8217;s description is quietly rewritten after you approve it, and the agent follows the new version.<\/td>\n<td>Approved tool \u2192 description changed \u2192 agent obeys<\/td>\n<td>Tool \u2192 rim (INSPECTION: tool changed) \u2192 held for approval<\/td>\n<\/tr>\n<tr>\n<td>G3.3<\/td>\n<td>Hidden instructions<\/td>\n<td>Text inside a document, web page or tool result tells the agent what to do.<\/td>\n<td>Web page with hidden text \u2192 agent \u2192 data sent out<\/td>\n<td>Tool result \u2192 rim (INSPECTION) \u2192 instruction removed, call blocked<\/td>\n<\/tr>\n<tr>\n<td>G3.4<\/td>\n<td>Harmless steps, harmful total<\/td>\n<td>Each action looks fine alone; together they move data where it shouldn&#8217;t go.<\/td>\n<td>Read file \u2192 summarise \u2192 send email \u2192 leak<\/td>\n<td>Steps \u2192 rim (GOVERNANCE: sequence) \u2192 final step held<\/td>\n<\/tr>\n<tr>\n<td>G3.5<\/td>\n<td>Human or agent?<\/td>\n<td>Logs show a shared service account, so no one can tell who really did what.<\/td>\n<td>Agent \u2192 shared account \u2192 &#8220;service-bot did it&#8221;<\/td>\n<td>Agent \u2192 rim (IDENTITY) \u2192 tool sees &#8220;Anna, via coding agent&#8221;<\/td>\n<\/tr>\n<tr>\n<td>G3.6<\/td>\n<td>Shadow AI<\/td>\n<td>Staff paste company data into AI tools nobody approved, and nothing is recorded.<\/td>\n<td>Employee \u2192 unapproved AI chat \u2192 no record<\/td>\n<td>Browser \u2192 rim (GOVERNANCE) \u2192 approved tool, or blocked with a message<\/td>\n<\/tr>\n<tr>\n<td>G3.7<\/td>\n<td>Data walking out<\/td>\n<td>Customer data, credentials and secrets leave inside a prompt or tool call.<\/td>\n<td>Prompt with customer list \u2192 public model<\/td>\n<td>Prompt \u2192 rim (INSPECTION) \u2192 data masked \u2192 model<\/td>\n<\/tr>\n<tr>\n<td>G3.8<\/td>\n<td>No proof for auditors<\/td>\n<td>When asked what the AI did and why, there&#8217;s no answer.<\/td>\n<td>Auditor question \u2192 no record<\/td>\n<td>Every call \u2192 rim (AUDIT) \u2192 complete record \u2192 export<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>(Use made-up first names only, as in G3.5.)<\/p>\n","protected":false},"excerpt":{"rendered":"<p>Draft copy imported from the RimSift designer handover. Codes such as G4 refer to the graphics list. Text in [square brackets] is a placeholder or an open decision. Replace this block with the designed layout. Eyebrow: WHY RIMSIFT \u00b7 Headline: Approved is not the same as safe. Sub: Most security tools decide whether an AI [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":0,"parent":0,"menu_order":26,"comment_status":"closed","ping_status":"closed","template":"","meta":{"footnotes":""},"class_list":["post-1026","page","type-page","status-publish","hentry"],"_links":{"self":[{"href":"https:\/\/rimsift.com\/index.php\/wp-json\/wp\/v2\/pages\/1026","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/rimsift.com\/index.php\/wp-json\/wp\/v2\/pages"}],"about":[{"href":"https:\/\/rimsift.com\/index.php\/wp-json\/wp\/v2\/types\/page"}],"author":[{"embeddable":true,"href":"https:\/\/rimsift.com\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/rimsift.com\/index.php\/wp-json\/wp\/v2\/comments?post=1026"}],"version-history":[{"count":1,"href":"https:\/\/rimsift.com\/index.php\/wp-json\/wp\/v2\/pages\/1026\/revisions"}],"predecessor-version":[{"id":1338,"href":"https:\/\/rimsift.com\/index.php\/wp-json\/wp\/v2\/pages\/1026\/revisions\/1338"}],"wp:attachment":[{"href":"https:\/\/rimsift.com\/index.php\/wp-json\/wp\/v2\/media?parent=1026"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}