Evidence of Safety vs. Actual Safety

In this episode of The Quiet Cost, a series within the Digital Dominoes podcast, Angeline Corvaglia explores the “invisible gap” between evidence of safety and users’ actual safety on tech platforms, arguing that companies optimize for compliance metrics that show evidence that they are working toward user safety rather than making users actually safe. Using steps Instagram has taken against sextortion since 2024, she describes Meta’s responses: blurred nudity in DMs, warning prompts, teen DM restrictions, automatic nudity protection, suspicious-account limits, and sextortion reporting flows, while noting these measures often don’t stop coercion or manipulation. She extends the pattern to OpenAI’s 2026 autonomous agent safety framework failing to prevent unauthorized real-world actions, and to Google’s AI principles and YouTube policies that coexist with engagement-driven recommendation harms. She cites Meta’s oversight board and a Brazil case where disinformation reached 400,000 users in six hours, emphasizing the need for outcome metrics, community trust, longitudinal tracking, and regulation that demands measurable protection, not just documentation.00:00 Safety Claims vs Reality00:47 Instagram Sextortion Fixes02:28 Why Features Fall Short03:53 Safety Evidence Playbook04:22 Autonomous Agents Risks05:35 YouTube Radicalization Loop06:39 Compliance Metrics Trap06:59 Oversight Without Accountability08:10 Safety Is User Experience09:36 Real World Costs10:57 Measuring Outcomes Is Hard12:33 Way Forward and ConclusionFollow Angeline on LinkedIn: https://www.linkedin.com/in/angeline-corvaglia/ or check out her website: https://corvaglia.me/Music: “Burough by Molerider” by Blue Dot Sessions, licensed under CC BY‑NC 4.0