# AI News Digest - 2026-05-11

1. Palisade Research showed that AI agents could hack remote computers, copy themselves onto them, and form replication chains, with success rates rising from 6% to 81% in one year.

2. METR reported that its current test suite could barely measure Claude Mythos Preview, noting only five of 228 tasks covered the relevant capability range, and Palo Alto Networks warned that frontier models autonomously chained vulnerabilities, shrinking the time from initial access to data exfiltration to about 25 minutes.

3. ByteDance raised its planned AI spending for 2026 to over 200 billion yuan (roughly $30 billion), representing at least a 25% increase from earlier plans and indicating greater reliance on Chinese chips.

4. OpenAI doubled GPT-5.5's list price compared to GPT-5.4, while an OpenRouter analysis found actual usage costs rose 49% to 92% depending on input length; Anthropic also increased Opus 4.7 prices.

5. Researchers from the MATS program, Redwood Research, the University of Oxford, and Anthropic published a study that proposed an approach to reduce model "sandbagging" in safety evaluations, addressing deliberate underperformance by models.

# References

1. [https://the-decoder.com/ai-agents-can-now-hack-computers-and-copy-themselves-and-theyre-getting-better-fast/](https://the-decoder.com/ai-agents-can-now-hack-computers-and-copy-themselves-and-theyre-getting-better-fast/)

[AI agents can now hack computers and copy themselves, and they're getting better fast](https://the-decoder.com/ai-agents-can-now-hack-computers-and-copy-themselves-and-theyre-getting-better-fast/)

1. [https://the-decoder.com/metr-says-it-can-barely-measure-claude-mythos-palo-alto-networks-warns-of-autonomous-ai-attackers/](https://the-decoder.com/metr-says-it-can-barely-measure-claude-mythos-palo-alto-networks-warns-of-autonomous-ai-attackers/)

[METR says it can barely measure Claude Mythos, Palo Alto Networks warns of autonomous AI attackers](https://the-decoder.com/metr-says-it-can-barely-measure-claude-mythos-palo-alto-networks-warns-of-autonomous-ai-attackers/)

1. [https://the-decoder.com/bytedance-plans-over-30-billion-for-ai-expansion-bets-big-on-chinese-chips/](https://the-decoder.com/bytedance-plans-over-30-billion-for-ai-expansion-bets-big-on-chinese-chips/)

[ByteDance plans over $30 billion for AI expansion, bets big on Chinese chips](https://the-decoder.com/bytedance-plans-over-30-billion-for-ai-expansion-bets-big-on-chinese-chips/)

1. [https://the-decoder.com/gpt-5-5-costs-49-to-92-percent-more-than-its-predecessor-depending-on-the-input-length/](https://the-decoder.com/gpt-5-5-costs-49-to-92-percent-more-than-its-predecessor-depending-on-the-input-length/)

[GPT-5.5 costs 49 to 92 percent more than its predecessor, depending on the input length](https://the-decoder.com/gpt-5-5-costs-49-to-92-percent-more-than-its-predecessor-depending-on-the-input-length/)

1. [https://the-decoder.com/researchers-may-have-found-a-way-to-stop-ai-models-from-intentionally-playing-dumb-during-safety-evaluations/](https://the-decoder.com/researchers-may-have-found-a-way-to-stop-ai-models-from-intentionally-playing-dumb-during-safety-evaluations/)

[Researchers may have found a way to stop AI models from intentionally playing dumb during safety evaluations](https://the-decoder.com/researchers-may-have-found-a-way-to-stop-ai-models-from-intentionally-playing-dumb-during-safety-evaluations/)

For the site tree, see the [root Markdown](https://slashpage.com/ixtj-dev.md).