[ctx]ctxseccontext security: attacks on the harness and context layer of AI agents, plus exploited-now news, hunting and playbooks.

← Threat Hunting · markdown

confirmed research (Push Security, Nov 5); detections from late October 2026

Hunt: fake OpenAI, SpaceX and Anthropic event RSVP pages chain Kali365 device-code phishing with VoidProxy AiTM to steal Microsoft and Google sessions

  • Delivery tells: invites via Google short links (c.gle / share.google), Zapier zpr.io, or an x.com logout redirect; Cloudflare 'Confirm you're not a robot' before the RSVP page; work account required to claim AI-tool rewards.
  • Lure hosts (Oct–Nov 2026 sample): eventrsvp[.]app (openai. / spacex. subdomains), rsvpconnect-events[.]com (anthropic.), path pattern /event/id<11 digits>/rsvp/ (also /live/, /join/); collection eventrsvp-217-60-102-48.sslip[.]io/collect.php (217.60.102[.]48).
  • PhaaS / proxy hosts used by this campaign: Kali365 backend 62zr4vcasl.shifts[.]icu; VoidProxy Google claudewebinnar[.]com (continue-to-event.); VoidProxy Microsoft continue-to-office-.ggleventapi[.]icu. These PhaaS hosts also serve other customers.
  • Defend: warn or block unexpected Microsoft device-code approvals (especially after marketing 'event' links); treat AI-branded RSVP pages that demand corporate SSO as phishing until proven; prefer Conditional Access that restricts device-code flow (see our device-code tip); hunt mail and proxy logs for the domains above and for pop-ups to login.microsoftonline.com/common/oauth2/deviceauth right after an RSVP click.

Sources: Push Security: Analyzing a fake event campaign combining AiTM and device code phishing payloads (Nov 5) · BleepingComputer: Fake OpenAI, SpaceX events used to steal Microsoft, Google accounts (Nov 5)