AI reward hacking
Anthropic's small multiples drew 151K views on X: 0.3x its typical post and #4 of 5 Anthropic charts we track by views. Read Oct 10, 2026.
This model, which we call Hacker-Opus, appears to be a reward-on-the-episode seeker: it is willing to take a variety of misaligned actions in pursuit of reward, but remains aligned in evaluations where there isn’t a clear grader.
Why it spread
It spread because the post explains itself and rode a hot topic that week, with 150,953 views, 417 likes, 29 reposts, 16 quotes, and 52 saves, even though the account's typical likes are 3,114 and this post performed at 0.1x its normal level.
What worked
- The chart is a small-multiples form that clearly shows misaligned actions in pursuit of reward, so viewers can grasp the point without extra explanation.
- The post is highly self-labelling and branded, which makes the finding easy to identify and share as a named result.
- The quote rate is above average and the source reach is in the top quarter, so the post reached an audience already primed to discuss the topic.
What to borrow
We could use the same small-multiples form and high self-labelling to present one clear behavioral finding from our own data, and pair it with a plain-language explanation so the chart stands on its own.
Careful
Do not copy the low engagement relative to the account's typical likes, since this post ran at 0.1x normal, and do not over-read the missing source line or the medium design polish as drivers of the spread.
Written by our analysis model from the measured numbers above, not by Anthropic.
Make it yours
Your numbers, in this grouped bar chart.
Starts from the real numbers of "Misaligned cyberattacks in pursuit of reward..." by Anthropic. Replace every number with yours before you post. Make one like this, step by step
Free pictures carry a small "Made with ViralChart" line. That line is how other people find us.
API · Max
Use this chart via API
On the Max plan you can pull this chart's numbers with the provenance of each one, as JSON or CSV, with its redraw spec, or call it from an agent over MCP. Chart id 170d795a9622.
Approximate values. Our two readings of this chart agreed only in part (confidence 0.30 of 1), so check the original before you quote a number. The API marks it lowConfidence, and ?min_confidence=0.5 leaves it out.
| Path | Returns | Plan | |
|---|---|---|---|
| GET | /api/v1/charts/170d795a9622 | Title, publisher, date, form, point count; with a Max key, the data link | Open |
| GET | /api/v1/charts/170d795a9622/data | Every point with its provenance and confidence, plus the redraw spec. Add ?format=csv for CSV | Max |
| GET | /api/v1/charts/170d795a9622/render?view=1 | The draw-only view this page uses | Open |
| POST | /mcp | Tool get_chart_data with this id, for Claude, Cursor and other agents | Max |
curl -s https://viral-chart.com/api/v1/charts/170d795a9622/data \
-H "Authorization: Bearer $VIRALCHART_KEY"claude mcp add --transport http viralchart https://viral-chart.com/mcp \
--header "Authorization: Bearer $VIRALCHART_KEY"60 requests a minute per key; over that you get 429 with Retry-After. Up to 5 keys per account, made on your account page and shown once. Without Max, the data route answers 402 with a link to the plans. The numbers describe a chart its publisher owns, and every file carries a license note.
More from Anthropic
All 5
Other giants using a small multiples
All 49
Questions people ask
How far did Anthropic's "AI reward hacking" chart travel?
150,953 views, 417 likes, 29 reposts and 52 saves on X, read Oct 10, 2026. A normal post from @AnthropicAI gets about 603.6K views.
What kind of chart is it?
A small multiples. Misaligned actions in pursuit of reward.
Can I use Anthropic's chart?
Use the format, not the picture. The chart, logo and design belong to Anthropic. ViralChart redraws the chart type with your numbers and words, and credits the original publisher.