September 2026, GASA Scam Fighter Award nomination and a Stanford workshop on multilingual trust and safety
Recent Updates From SimPPL
Last month, SimPPL was nominated for the Global Anti-Scam Alliance (GASA) Scam Fighter Award for our work with Migrasia, using Arbiter to uncover 14+ networks promoting potential job scams targeting migrant workers in Southeast Asia.
SimPPL has also been selected to lead a workshop at Stanford University on cross-platform investigations and multilingual trust and safety using Arbiter. The session will explore how investigators can examine harms spanning multiple platforms, languages, and communities, drawing on work already being conducted with Arbiter across 20 countries. If you or someone you know is around, we would love to say hi. Join our early partners in developing investigations that we will feature in our workshop!
Investigations Conducted Using Arbiter
Scam Ecosystem: What Is Driving the Conversation
Recent scam cases in India show how investment fraud is being packaged in different ways, from apps and betting offers to impersonation schemes. Many of these schemes use Telegram to promote offers or move users into private groups, so Arbiter analysed 1,005,370 messages from 1,032 public channels to see what was happening there.
What Arbiter Found
- Some channels showed users claiming large returns from small investments. One claimed to have made around $20,000 after investing about $1,500.
- Betting channels promised users money back even if they lost. One promotion offered to return 50% of a losing bet.
- Verified accounts were also being bought and sold. Identity-verified accounts were advertised for $5–$50, alongside offers for card and bank data.
MMDR Amendment Act: What Is Driving the Conversation
India passed the MMDR Amendment Act in August, removing states' ability to levy a separate tax on mineral rights. The change sparked debate over its impact on state revenue and taxing powers.
What Arbiter Found
- Criticism began as soon as the Bill was introduced and reached 8.29M engagements. Critics argued that states would lose taxing powers and revenue, and that pending dues from mining companies could no longer be collected.
- The case for the amendment came largely after Parliament had passed it. Only 13 supporting posts appeared before 13 August. Supporters later argued that states still receive 90% of mining revenue and that a unified tax regime would improve transparency.
- The fight is now moving from social media to the courts. The amendment makes uncollected past mining dues unrecoverable, and states are now planning to challenge the change in the Supreme Court.
AI Safety Risks and Alarm Signals
Warnings from OpenAI and Bill Gates triggered another round of debate about advanced AI risks in August. Gates warned about job losses, cyberattacks and biological threats as AI capabilities advance. But the conversation quickly shifted from those risks to the credibility and influence of the people raising them.
What Arbiter Found
- The debate centered on how dangerous increasingly capable AI systems could become. Supporters of stronger safeguards pointed to unintended model behaviour and risks such as job losses, cyberattacks and biological threats. Critics, meanwhile, questioned the people and organisations sounding these alarms and what they stood to gain.
- Gates's warning about AI risks became controversial itself. Some posts treated the fact that Gates was sounding the alarm as a reason to take the risks seriously. Others criticised him for warning about AI capabilities that his own foundation had helped fund.
- The policy debate received far less attention than the warnings themselves. California's AI safety bill, SB 53, drew little engagement, while posts about broader AI risks received thousands of likes.
- Headlong: An Open-Source Agent Microharness with Persistent Agency and Recursive LLMs
Agents that run continuous self-guided thinking loops instead of waiting for prompts. Built on Bash and recursive LLMs, they execute shell commands to manage projects and work across Slack and Telegram.
- Transformer Explainer: LLM Transformer Model Visually Explained
A browser tool that shows GPT-2 predicting the next token, layer by layer: embeddings, multi-head attention, MLP layers and output probabilities, updating in real time.
- AI and Democracy: Mapping Notions, Debates and Challenges (Philosophy & Technology)
A map of where the AI-and-democracy debate sits, treating AI as both a threat to democratic processes and a tool for civic engagement.
- Measuring Narrative Polarization in Online Discourse (PNAS Nexus)
Uses LLMs to measure polarization in narratives rather than users or networks. On YouTube content about the Israeli–Palestinian conflict, creators push sharply opposed narratives while commenters converge on shared story structures.