Latest / ELECTE's Podcast: AI Frontiers / The Person Is Real. The Deception Is Too.

Transcript

The Person Is Real. The Deception Is Too. · ELECTE's Podcast: AI Frontiers

• 0:03
  SPEAKER_00: This is AI Frontiers.
• 0:06
  SPEAKER_00: Today, why real people make fake stories convincing?
• 0:10
  SPEAKER_00: A genuine signal proves far less than it seems.
• 0:14
  SPEAKER_00: The mistake is granting one true piece of evidence authority over a story it cannot prove.
• 0:19
  SPEAKER_00: Anthropics' September 2026 Threat Report describes a deceptive dating app network that combined artificial intelligence personas with paid human workers.
• 0:29
  SPEAKER_00: Over two weeks in April, it observed more than 4,700 AI personas interacting with at least 25,000 people.
• 0:38
  SPEAKER_00: Real workers supplied live video calls and social media followbacks that helped reassure users.
• 0:45
  SPEAKER_00: Even those workers were AI assisted.
• 0:47
  SPEAKER_00: A separate model proposed their replies and they tapped one.
• 0:51
  SPEAKER_00: These are one provider's findings from one operation, not market-wide estimates.
• 0:56
  SPEAKER_00: The work can be real too.
• 0:58
  SPEAKER_00: Microsoft's March 2026 research describes North Korean information technology worker operations using AI to sustain everyday correspondence, technical responses, and coding work under fraudulent identities.
• 1:12
  SPEAKER_00: Microsoft says AI helps actors, quote, get hired, stay hired, and misuse access at scale.
• 1:20
  SPEAKER_00: In the Christina Chapman case, the United States Department of Justice reported 309 U.S.
• 1:26
  SPEAKER_00: businesses defrauded and more than$17 million generated by a remote worker scheme supported by stolen identities and a domestic laptop farm.
• 1:35
  SPEAKER_00: It ran from 2020 to 2023.
• 1:38
  SPEAKER_00: Those figures are not a measure of losses caused by generative AI.
• 1:42
  SPEAKER_00: Agreement between sources is only worth as much as their independence.
• 1:46
  SPEAKER_00: Four documents supplied by the same party are not four confirmations.
• 1:51
  SPEAKER_00: A reference reached through contact details obtained from the former employer tests a different relationship from one reached through the applicant's preferred address.
• 1:59
  SPEAKER_00: Detection still matters, but its result has a boundary.
• 2:03
  SPEAKER_00: A September 2026 preprint introducing DF26 found participants correctly identified 52.6% of its fake clips, compared with 74.5% on CLib DF and 69.8% on DeepSpeak V2.
• 2:21
  SPEAKER_00: Five-second clips, no audio, mostly technical university volunteers.
• 2:26
  SPEAKER_00: A separate March 2026 preprint found mean human accuracy of 78.4% versus 53.7% for the tested detector variants.
• 2:37
  SPEAKER_00: Different tasks, different samples.
• 2:40
  SPEAKER_00: There is no useful universal score.
• 2:43
  SPEAKER_00: Cloudflare's 2026 announcement with name tag adds identity verification to workplace access.
• 2:50
  SPEAKER_00: A selfie and government-issued identity document produce a token used in the access decision.
• 2:56
  SPEAKER_00: That is a workflow, not proof of honest purpose.
• 3:00
  SPEAKER_00: So ask one question in every hiring review or access decision.
• 3:05
  SPEAKER_00: What would still be unproven if every item in front of us were genuine?
• 3:08
  SPEAKER_00: Then narrow the access or contact the organization yourself.
• 3:12
  SPEAKER_00: That's AI Frontiers.