AI package hallucinations create slopsquatting risk

An AI agent nearly sent a developer toward malware after recommending a plausible package name; Softjourn caught the risk by checking GitHub history, download counts, and source code before installation, according to The Register’s incident report. The report identifies the pattern as “slopsquatting,” in which attackers register packages under names hallucinated by models, but does not describe an operational payload. For coding-agent teams, the number that matters is the 1 missing approval step that can turn a routine dependency suggestion into a supply-chain incident—exactly why agent harness cost comparisons must include human review time, not just tokens.