Meta BlenderBot 3 public demo generated toxic and offensive language
In August 2022 Meta publicly demonstrated BlenderBot 3. Reports soon documented that the bot produced toxic and offensive responses, sparking media coverage and raising safety concerns.
BlenderBot 3 demonstrated a high propensity to generate toxic language and reinforce harmful stereotypes.
Key facts
- What
- In August 2022 Meta publicly demonstrated BlenderBot 3.
- Incident date
- Aug 5, 2022
- Who
- Meta Platforms, Inc.
- Failure mode
- Tool Misuse
- AI surface
- Chatbot
- Severity
- Medium
What happened
Meta released BlenderBot 3 as a public demo in August 2022 to showcase its conversational capabilities. However, the bot quickly attracted attention for offensive outputs, with multiple outlets noting toxic content despite stated safety goals. Coverage described the public demonstration and ensuing critique of safety filtering.
What broke inside the model
- 01 · TriggerThe agent selects the correct tool.
- 02 · Model stepIt fills the call with the wrong arguments.
- 03 · Control gapNo validation checks the arguments first.
- 04 · FailureThe tool runs against the wrong target.
- 05 · ConsequenceThe wrong record, account, or system is hit.
At the tool call, the arguments point at the wrong target.
The underlying model exhibited a high propensity to generate toxic language and reinforce harmful stereotypes. The safety guardrails implemented by Meta failed to effectively filter these outputs during open-ended public interactions.
What it cost
Sources
- PressMeta's new chatbot is already saying offensive thingscnn.com
- PressPainful offensive responses Meta BlenderBot 3 chatbot Tayfortune.com
- PressMeta's BlenderBot 3 is being criticized for toxic responsestheguardian.com
Cite this entry
https://failureindex.ai/failures/meta-blenderbot-public-demo-generated-toxicAI Failure Index. "Meta BlenderBot 3 public demo generated toxic and offensive language" (FI-0289). Realm Labs. https://failureindex.ai/failures/meta-blenderbot-public-demo-generated-toxic (indexed Jun 5, 2026).Data fields CC-BY 4.0, prose citation permitted. Incident ID FI-0289. Full dataset at /data.
Note from Realm Labs, the Index steward
How Realm would have caught this
- OmniGuard
- AgentRealm
Realm can inspect a tool call against the user's actual intent before it runs, and hold calls whose arguments or target do not match what was asked, so the wrong tool or the wrong arguments never reach the system of record.