"Anthropic states that it has safeguards in place to prevent common risks like prompt injection and will limit access to certain 'off limits' apps by default, such as investment and trading platforms."
"The model is trained to avoid risky operations, including moving or investing money, modifying files, scraping facial images, or inputting sensitive data, but these safeguards are not absolute."
"When computer use is activated, Claude will be able to see anything visible on-screen, including personal data, sensitive documents, or private information, leading to a recommendation to avoid working with sensitive data."
Anthropic has established safeguards to mitigate risks such as prompt injection and restrict access to certain applications, including investment platforms. The AI model is trained to avoid risky operations like financial transactions and handling sensitive data. However, the company admits that these safeguards are not foolproof, and the AI may occasionally operate outside its intended boundaries. Additionally, when activated, the AI can access visible on-screen information, prompting the recommendation to avoid using sensitive data during the research preview phase.
Read at Ars Technica
Unable to calculate read time
Collection
[
|
...
]