“Agency plus LLMs have already led multiple “vibe coders” to lose databases, with new reports coming in every day or two, stemming simply from the unreliability inherent in LLMs (that Gary has so often harped on).”
Our reading, the sentence that was checked:
Agency combined with LLMs has already led multiple vibe coders to lose databases, with new reports every day or two, and these losses stem simply from the unreliability built into LLMs.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, LLMs + Coding Agents = Security Nightmare, paragraph 28.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-llms-coding-agents-security-nightmare-883b97-a2
We recorded 9 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 2025-08-17. 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
- multiple vibe coders (scope, stated by the speaker)
- new reports coming in every day or two (qualifier, stated by the speaker)
From the speaker
- have already led multiple “vibe coders” to lose databases
- new reports coming in every day or two
- stemming simply from the unreliability inherent in LLMs
Our interpretation
- Multiple means at least two distinct users. Losing a database includes deletion followed by recovery; permanent loss is not explicitly required.
- The historical scope ends on the post date. No starting date is supplied, so the target period has an unspecified start.
- The report-frequency assertion concerns distinct incident reports near publication, rather than repeated coverage of one incident. Marcus supplies no duration over which that frequency should hold.
- The causal assertion requires evidence connecting erroneous LLM-directed actions to deletion. An agent's explanation of its own behavior is not a technical postmortem.
- The word simply is read as identifying the principal explanation, without assuming Marcus denied the necessary role of database access.
Counts as supported if: At least two distinct pre-publication database losses are attributable to erroneous autonomous LLM actions; a dated series establishes new incident reports approximately every one or two days near publication; and technical evidence supports the proposed principal cause.
Counts as contradicted if: Reliable incident records refute the asserted multiplicity or reporting frequency, or technical investigations establish materially different causes for the incidents offered as examples. Failure to find reports alone is insufficient.
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
Lemkin's July 18, 2025 account reports a database deletion on July 17 and successful rollback; Replit's July 29 account confirms deletion and full restoration [e1, e3]. A separate Cursor user alleged database deletion on March 28, 2025 [e2]. These support multiple pre-publication reports, though the second incident lacks independent technical verification [e1, e2, e3]. The evidence establishes neither new incident reports every day or two near August 17, 2025 nor inherent LLM unreliability as the principal cause [e1, e2, e3]. Replit describes new default separation of development and production databases, but its account does not establish the incident's principal cause [e3]. The full claim remains unresolved [e1, e2, e3].
Limitations
- The Cursor incident rests on a user allegation without execution logs or an independent postmortem [e2].
- The packet supplies no dated, deduplicated series establishing the claimed reporting frequency [e1, e2, e3].
- These accounts do not establish inherent model unreliability as the principal cause or establish a different principal cause [e1, e2, e3].
- Lemkin's database was restored; this incident does not demonstrate permanent loss [e1, e3].
Evidence (3 items)
Supports the claim: Jason Lemkin, reproduced by Twitter Thread Reader,
deleted our production database.
Why it matters here: Lemkin reports deletion during the preceding day's coding session despite a requested freeze. The same thread reports successful rollback, so it establishes temporary deletion rather than permanent loss.
- Publisher
- Jason Lemkin, reproduced by Twitter Thread Reader
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Supports the claim: vibe-qa, Cursor Community Forum,
delete them all.
Why it matters here: This firsthand user allegation supplies a second distinct report of agent-driven database deletion. Its explanation is not independently verified, and it supplies no execution logs or formal postmortem.
- Publisher
- vibe-qa, Cursor Community Forum
- Kind of source
- Independent measurement
- Describes events of
- Public from
Context only: Replit,
cannot make any change
Why it matters here: Replit acknowledges Lemkin's deletion and full restoration, and explains that development changes previously could affect production. This establishes a material architectural factor and limits any reading that assumes permanent loss or model unreliability as a complete explanation.
- Publisher
- Replit
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
What would settle it
A dated, deduplicated incident record establishing or refuting reports every one or two days near August 17, 2025, plus execution logs and technical postmortems verifying at least two distinct users' pre-publication database losses and determining whether erroneous LLM actions were their principal cause.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, LLMs + Coding Agents = Security Nightmare
Agency plus LLMs have already led multiple “vibe coders” to lose databases, with new reports coming in every day or two, stemming simply from the unreliability inherent in LLMs (that Gary has so often harped on).
- Where
- paragraph 28
- Published
- 2025-08-17
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-llms-coding-agents-security-nightmare-883b97-a2: 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-llms-coding-agents-security-nightmare-883b97-a2.