The 'local attacker' classification in CVE-2026-70310 is the most analytically significant detail, and it should change how you evaluate this vulnerability regardless of its moderate CVSS 5.5 score.
Standard Word vulnerabilities assume remote delivery: an attacker sends a malicious document, the victim opens it, exploitation proceeds. This CVE explicitly states 'local attacker' — meaning an adversary who already has some execution capability on the target system and is using this OOB read as part of their operational toolkit. This reframes the vulnerability entirely. Either this is a defense-in-depth failure requiring a separate initial access vector, or Microsoft is quietly acknowledging a specific local attack surface in Word that deserves scrutiny: preview handlers, COM/OLE interfaces invoked programmatically, or file format parsers accessed by other applications.
The CVSS 5.5 understates the real risk. An out-of-bounds read in Word's parser can leak ASLR pointers, credential fragments, session tokens, or heap layout information — that's not a stepping stone to RCE, that's the prize for an attacker who already has local access. The historical pattern with Microsoft Office CVEs is telling: vulnerabilities flagged as 'local attacker' frequently get revised upward within 90 days when researchers discover the parser state being leaked is richer than initially disclosed. The CVSS accurately scores the isolated bug; it doesn't capture what an attacker with local foothold actually targets.
You should also ask a supply chain question, not just a patching question: what in your environment invokes Word programmatically — server-side document processing, backup workflows, monitoring pipelines, document management systems? Those integrations assume Word is a hardened execution context, but programmatic invocation changes the attack surface entirely. An attacker who can control what gets fed to Word through COM automation can craft inputs that exercise legacy parser branches for deprecated file formats that fuzzing never covers because those branches handle formats users no longer consciously invoke. This is legacy sediment — parser code maintained for backward compatibility that receives no adversarial testing because nobody remembers it exists.
The low EPSS (0.00352) likely reflects that active exploitation isn't public yet, not that the vulnerability is theoretical. The 'local attacker' flag suggests Microsoft's internal testing found this through audit of programmatic interfaces rather than document-based attack simulation — meaning this exists in a code path external exploit developers haven't charted. That makes it more dangerous, not less. Organizations treating this as a non-urgent finding because the CVSS is 5.5 are extending their exposure window precisely when they have the most to lose: an already-compromised machine with an unpatched information disclosure flaw is paying interest on two exposures simultaneously, and the patch queue treats them as independent findings when they're not.