Post
"Across multiple news-reported incidents, Australians — particularly small businesses, artists, political discussion groups, community organizations, and publishers — experienced sudden loss of Facebook and Instagram accounts, vague violation notices, inaccessible or ineffective appeals, and severe harm to livelihoods or community activity."
@mendicot, 2025
(1/?)
"The mechanism underlying these cases was largely automated: AI-driven moderation, impersonation sweeps, and group-management bots prone to misclassification, with little human review and templated notices that failed to specify the exact evidence or steps for remediation. This automation-first approach is compounded by the absence of EU-style safeguards in Australia, making 'no-recourse' outcomes more likely."
@mendicot, 2025
(2/?)
Welcome to Zion @mendicot! No DataFarming power plants or squiddys to mess with you here. The dance party starts in an hour :P
(3/?)
The fediverse doesn't really have a full replacement for FarceBook yet, but I suggest you check out @bonfire which has just hit a 1.0 release;
https://bonfirenetworks.org/posts/announcing-the-bonfire-social-1-0-release-candidate/
Older fediverse apps that have been compared to FB include Friendica and Hubzilla. You can explore more fediverse software here;
https://delightful.coding.social/delightful-fediverse-experience/
... and here too;
https://fediverse.party/en/miscellaneous/
But this one needs some updates. I was recently shoulder-tapped to take over as maintainer, and I'm working on it.
(4/?)
"Corrective measures identified in the research are precise and testable within administrative-law and audit frameworks."
@mendicot, 2025
(5/?)
"They include enforcement notices that cite exact rules and supporting evidence, guaranteed human-in-the-loop review for severe penalties, time-bound two-stage appeals with default restoration if deadlines lapse, localized risk controls adapted to Australian conditions, routine publication of country-level error, reversal, and appeal-success rates, and pre-deployment audits of classifiers and group-management bots against Australian data."
@mendicot, 2025
(6/?)
@mendicot
> guaranteed human-in-the-loop review for severe penalties
I'd go further, following the "centaurs" vs "reverse-centaurs" frameworks @pluralistic describes here;
https://pluralistic.net/2024/08/02/despotism-on-demand/#virtual-whips
All moderation tools must be built to augment the work of human moderators. Making moderation workers into the overseers of automated systems, which they don't direct and control, just sets them up to take the blame for the inevitable failures of a Trained #MOLE;
https://disintermedia.substack.com/p/invasion-of-the-mole-trainers
(7/7)
A space for Bonfire maintainers and contributors to communicate
