Google warns staff about using chatbots

Alphabet is cautioning employees about how they use chatbots, including its Bard, at the same time as it markets the program around the world, according to four people familiar with the matter.

The parent company of Google has advised staff not to enter its confidential materials into AI chatbots, the people said, which was confirmed by the company, citing long-standing policy on safeguarding information. The chatbots, among them Bard and ChatGPT, are human-sounding programs that use so-called generative artificial intelligence to hold conversations with users and answer myriad prompts.

Human reviewers may read the chats, and researchers found that similar AI could reproduce the data it absorbed during training, creating a leak risk.

Alphabet also alerted its engineers to avoid direct use of computer code that chatbots can generate, some of the people said. Asked for comment, the company said Bard can make undesired code suggestions, but it helps programmers nonetheless.

Google also said it aimed to be transparent about the limitations of its technology. The concerns indicate Google’s intention to avoid business harm from software it launched in competition with ChatGPT.

At stake in Google’s race against ChatGPT’s backers — OpenAI and Microsoft — are billions of dollars of investment and still untold advertising and cloud revenue from new AI programs. Google’s caution also reflects what’s becoming a security standard for corporations: warning personnel about using publicly-available chat programs.

A growing number of businesses have set up guardrails on AI chatbots, among them Samsung, Amazon.com and Deutsche Bank. Apple, which did not respond to requests for comment, reportedly has as well.

About 43% of professionals were using ChatGPT or other AI tools by January, often without telling their bosses, according to a survey of nearly 12,000 respondents, including from top US-based companies, by the networking site Fishbowl.

By February, Google told staff testing Bard before its launch not to give it internal information, Insider reported. Now Google is rolling out Bard to more than 180 countries and in 40 languages as a springboard for creativity, and its warnings extend to its code suggestions.

Google said it has had detailed conversations with Ireland’s Data Protection Commission and is addressing regulators’ questions, after a Politico report on Tuesday that the company was postponing Bard’s EU launch this week pending more information about the chatbot’s impact on privacy.

Sensitive information

Such technology can draft emails, documents, even software, promising to vastly speed up tasks. Included in this content, however, can be misinformation, sensitive data or even copyrighted passages from a Harry Potter novel.

A Google privacy notice updated on June 1 states: “Don’t include confidential or sensitive information in your Bard conversations.”

Some companies have developed software to address such concerns. Cloudflare, which defends websites against cyberattacks and offers other cloud services, is marketing a capability for businesses to tag and restrict some data from flowing externally.

Google and Microsoft also are offering conversational tools to business customers that will come with a higher price tag but refrain from absorbing data into public AI models. The default setting in Bard and ChatGPT is to save users’ conversation history, which users can opt to delete.

It “makes sense” that companies would not want their staff to use public chatbots for work, said Yusuf Mehdi, Microsoft’s consumer chief marketing officer. “Companies are taking a duly conservative standpoint,” Mehdi said of a comparison between the company’s free Bing chatbot and its enterprise software. “There, our policies are much more strict.” 

Microsoft declined to comment on whether it has a blanket ban on staff entering confidential information into public AI programs, including its own, though another executive there said he personally restricted his use.

Cloudflare CEO Matthew Prince said that typing confidential matters into chatbots was like “turning a bunch of PhD students loose in all of your private records”.

Reuters