liveBook · Manning

Tool chaining

As explained in AI Governance

Tool chaining is the composition of multiple tool operations into a sequence to achieve a task or produce an effect.

Tool chaining combines legitimate tool operations into a sequence that can produce a harmful effect. In adversarial testing, it probes whether permitted tools become dangerous when composed across multi-step interactions. section 4.8.8

Red-team test

Test whether an attacker can combine legitimate tools into a harmful sequence, rather than assessing each tool only in isolation. section 4.8.8

How AI Governance covers it

How deep it goes
Defines the attack pattern and places it in adversarial testing, but gives no concrete chains or implementation procedure.
Examples
Tests harmful sequences formed by combining legitimate tools, without specifying a particular tool sequence.
What you can do after
Evaluate agent exposure by testing whether legitimate tools can be combined into harmful sequences.
How it connects
Links tool chaining to red teaming, multi-turn attacks, memory poisoning, and goal manipulation.

Also explained in

Compare how 4 books explain tool chaining →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of AI Governance. The section links point to the parts of the book it draws on.