“Although it is unreliable and prone to hallucinations, it is terrific at making plausible sounding text (e.g., phishing attacks to trick people into revealing credentials) and deepfaked videos at virtually zero cost, allowing attackers to broaden their attacks.”
Our reading, the sentence that was checked:
Although generative AI is unreliable and prone to hallucinations, it is terrific at making plausible sounding text (such as phishing attacks) and deepfaked videos at virtually zero cost, allowing attackers to broaden their attacks.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Breaking news: U.S. banks are taking hits from AI agents, paragraph 13.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-breaking-news-us-banks-are-taking-055d32-a1
We recorded 12 claims from this post. See every claim from this post.
The wording above is the version this site observed on 2026-10-28. The source may have been edited after 2026-10-22. This page does not claim the wording was captured then.
What the claim commits to
- Kind
- Causal claim
- Deadline, as the speaker wrote it
- None stated. This site does not add one.
- Deadline state when assessed
- No horizon
- Deadline state on 2026-10-29
- No horizon (the release’s reference date)
Conditions
- restated from his early-2025 Politico warning (qualifier, stated by the speaker)
From the speaker
- unreliable and prone to hallucinations
- terrific at making plausible sounding text
- phishing attacks
- deepfaked videos
- at virtually zero cost
- allowing attackers to broaden their attacks
Our interpretation
- The ordinary reading is that generative AI can produce convincing phishing text and deceptive video cheaply enough to substantially reduce the effort needed to reach more victims.
- Virtually zero cost means negligible marginal production expense, not literally free hardware, labor, distribution, or an entire criminal operation.
- The cost wording applies to both text and video. Required generation fees, compute, and repeated attempts count toward marginal production expense.
- The passage restates an early-2025 warning. Evidence must establish capabilities available by then; it is not a forecast with a future deadline.
Counts as supported if: Capability evaluations, production-cost evidence, and documented attacker use jointly establish convincing text and video at negligible marginal cost, with reduced effort enabling broader attacks despite reliability failures.
Counts as contradicted if: Representative evidence establishes that a central component fails, such as effective deceptive video requiring materially non-negligible marginal production costs under the stated reading.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Does the evidence support the cause?
Outcome: Unresolved
The record was checked, and the evidence available by the cutoff does not settle it.
Scored on a plain reading. The speaker named no measure for this claim, so it was judged on the reading an informed reader would take, stated below under our interpretation. Both reviews had to find that reading fair.
- Evidence as of
- Deadline state when assessed
- No horizon
- Assessment recorded
- Policy version
- 2026-10-28.7
Rationale
The phishing experiment and FBI warning support convincing AI-generated deception and reduced effort that broadens attacks. Unit 42 also documents real video scams. The stronger cost claim remains unsettled: commercial video services charged meaningful amounts, but their prices do not measure marginal self-generation costs. That mismatch prevents either supporting the whole statement or treating those prices as a decisive contradiction.
Limitations
- Virtually zero is undefined.
- The phishing experiment measures clicks, not completed credential theft.
- The FBI source is a qualitative official warning, not a controlled cost study.
Evidence (3 items)
Supports the claim: Heiding and colleagues, arXiv,
on par with human experts
Why it matters here: The 101-participant evaluation found comparable click-through rates for automated AI phishing and expert-written messages, while its economic analysis found advantages from scaling automation. This supports convincing text and lower costs, but does not measure deceptive-video production.
- Where
- Abstract
- Publisher
- Heiding and colleagues, arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Supports the claim: Federal Bureau of Investigation, Internet Crime Complaint Center,
reduces the time and effort
Why it matters here: The FBI describes faster production of believable messages, wider audiences, and deceptive video chats used in fraud. This supports the proposed scaling mechanism and actual use of both media, but supplies no numerical production-cost estimate.
- Where
- Opening paragraph
- Publisher
- Federal Bureau of Investigation, Internet Crime Complaint Center
- Kind of source
- Official statistics
- Describes events of
- Public from
Context only: Palo Alto Networks Unit 42,
manipulate visuals and audio
Why it matters here: Unit 42 documents deployed deepfake scam campaigns and quotes typical video-production prices of $60–$500. Those service prices caution against describing all production as virtually free, but they do not measure the marginal cost of self-generation and therefore do not decisively contradict that reading.
- Publisher
- Palo Alto Networks Unit 42
- Kind of source
- Independent measurement
- Describes events of
- Public from
What would settle it
Contemporaneous tests of convincing phishing text and impersonation video that report successful-output rates, required generation fees and compute, and costs including retries would settle the missing cost component.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Breaking news: U.S. banks are taking hits from AI agents
Although it is unreliable and prone to hallucinations, it is terrific at making plausible sounding text (e.g., phishing attacks to trick people into revealing credentials) and deepfaked videos at virtually zero cost, allowing attackers to broaden their attacks.
- Where
- paragraph 13
- Published
- 2026-10-22
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-breaking-news-us-banks-are-taking-055d32-a1: Current. Outcome: Unresolved. Evidence as of 2026-10-28. Public date not recorded.
Recorded events
Assessment publishedBy this site
First assessment published after both reviews and the final-text check passed.
About assessment marcus-breaking-news-us-banks-are-taking-055d32-a1.