Skip to content

Security update: React 19.0.3 - #3

Merged
MaryWylde merged 1 commit into
devfrom
chore/bump-react-19.0.3
Dec 12, 2025
Merged

MaryWylde merged 1 commit into
devfrom
chore/bump-react-19.0.3

Conversation

@MaryWylde

Copy link
Copy Markdown
Contributor

No description provided.

@MaryWylde
MaryWylde merged commit 45eb04c into dev Dec 12, 2025
1 check passed
manager added a commit that referenced this pull request May 21, 2026
…usory truth)

Two new entries that follow the same structure as availability-heuristics:
scenario, two side-by-side email-style cards (with the bias-exploiting
side flagged), why-it-works paragraph, and a four-move Protect-yourself
list keyed to the bias mechanism.

- Attentional bias: a loud "URGENT sign-in from Moscow" decoy beside a
  quiet "updated wire details" ask one minute earlier. The bias mechanism
  is attention budget — the noisy one is bait, the quiet one is the
  attack. Defenses: scan the same window after the loud event, never
  trust an alert's link, decoys travel in pairs.
- Illusory truth effect: a cold "vendor banking update" ask vs. the
  same content reframed as a third touch ("Re: Re: ... as mentioned
  last week"). Two prior friendly notes were a deposit into your
  credibility account; the third is the withdrawal. Defenses: thread
  length is not verification, treat first money/credential ask as new
  no matter how familiar, cross-check sender against external contacts.

Plus a small refactor: OffsecBiasContent moved out of
availabilityHeuristics.ts into a dedicated types.ts so the type now
defines attachment/flagged as truly optional and the same shape applies
across all 105 future biases without each file needing to mirror every
field of the first one.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
manager added a commit that referenced this pull request May 21, 2026
…chat DMs

Email was overused for the first three examples. Generalised the visual
chassis with a discriminated card kind so each bias picks the surface
that matches its actual attack vector. Email stays for availability
(news-anchored phishing IS email-shaped); attentional bias #2 pivots to
two competing push notifications (Microsoft Defender "unauthorised
sign-in" decoy vs Wire Approvals "bank details changed" quiet ask);
illusory truth #3 pivots to a LinkedIn-style chat (cold ask vs same
message reframed as the third touch with a "2 messages this month"
context line).

Future biases (browser alerts, OS dialogs, voice calls, etc.) can add
their own card kind to types.ts without breaking the existing two.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant