Women have long had to strike a Goldilocks balance at work — not too harsh, not too soft. Now some are using AI to help get the tone just right.

Why it matters: Smart chatbot use is more of a shortcut to clear communication than a solution to the underlying problem of workplace gender bias.

Women have faced judgment for years on their physical appearance at work.

• Now that so much of work takes place in inboxes, Slack and Teams messages, that burden has shifted form — and text can be just as fraught.

What they're saying: "I come from the U.K., so my style of communication is sometimes very different from how Californians communicate," Alice Chan, a San Francisco marketing professional, told Axios. "I have been told sometimes that I'm too direct."

• Chan says she still writes emails and other messages herself, but then drops the text into ChatGPT and asks how it might be received.

"I love using AI for vendor negotiations," Susannah Shattuck, head of product for Credo AI, told Axios in an email. "I'll ask Claude to review and revise my counteroffer — and almost always, I find that it pushes me to be a tougher negotiator."

• "I want to be able to set my tone," Angela Tran, account executive at Astrsk PR, told Axios. Tran says she runs her emails and sometimes even her Slack messages through AI. "I don't want to come off as too pushy," she says.
• Tran also uses Grammarly's AI tone detector tool to determine how her messages might sound to others.

Zoom out: The emotional toll of self-editing at work is real. Women spend significant energy calibrating their tone, a burden less often felt by their male peers.

• "I wish I didn't have to do it, but it's working," consultant Jennifer Borchardt told Axios about running her communications through ChatGPT and Claude to help them sound more "conciliatory," but also more confident.
• Borchardt says that since she's been massaging her words with AI, she's getting faster, more positive and more constructive responses to her messages. 

By the numbers: Women say they feel pressured to use a "work voice," according to a survey from Preply, a corporate communications platform.

• 82% of respondents said they change their communication style for the workplace.
• One in six women said they've been told to change their tone at work.

Zoom in: Women say they're trying to find a middle ground that feels comfortable but also checks all the boxes of "invisible corporate expectations," Kate Mason, author of the new book "Powerfully Likeable: A Woman's Guide to Effective Communication," told Axios.

• "There's definitely an empowering aspect to having an in-your-pocket adviser like ChatGPT," Mason said.

The other side: Mason says women use AI in workplace communications for the same reason men do: to sound knowledgeable quickly. But she doesn't see it as a cure-all.

• "Anyone using AI as their primary mode of writing risks not only losing their own voice and sense of self in their communication, but also risks atrophying their own logic, insights and thinking," says Mason.
• "If you outsource all text to a machine, you're not learning how you might want to respond in real life, which I think is a lost opportunity to practice."

Between the lines: Sometimes the same message could be interpreted as negatively aggressive or positively assertive based on who's reading it, Borchardt says.

• She's been experimenting with what she calls "hyper-personalizing" her communications.
• She drafts emails, drops them into ChatGPT with the recipient's LinkedIn profile and bio, and asks for a rewrite tailored to that person.

What we're watching: Younger women might be bucking the trend of policing their own tones.

• "They have less deference and are less likely to apologize than their predecessors," Mason says.
• "My hope is that they will continue to find modes of communication which serve themselves and their colleagues, while at the same time, not having to perform outdated versions of femininity."

Google DeepMind said today it has updated a key AI safety document to account for new threats — including the risk that a frontier model might try to block humans from shutting it down or modifying it.

Why it matters: Some recent AI models have shown an ability, at least in test scenarios, to plot and even resort to deception to achieve their goals.

Driving the news: The latest Frontier Safety Framework also adds a new category for persuasiveness, to address models that could become so effective at persuasion that they're able to change users' beliefs.

• Google labels this risk "harmful manipulation," which it defines as "AI models with powerful manipulative capabilities that could be misused to systematically and substantially change beliefs and behaviors in identified high stakes contexts."
• Asked what actions Google is taking to limit such a danger, a Google DeepMind representative told Axios: "We continue to track this capability and have developed a new suite of evaluations which includes human participant studies to measure and test for [relevant] capabilities."

The big picture: Google DeepMind updates its Frontier Safety Framework at least annually to highlight new and emerging threats, which it labels "Critical Capability Levels."

• "These are capability levels at which, absent mitigation measures, frontier AI models or systems may pose heightened risk of severe harm," Google said.
• OpenAI has a similar "preparedness framework," introduced in 2023.

The intrigue: Earlier this year OpenAI removed "persuasiveness" as a specific risk category under which new models should be evaluated.

• AI investment is fueling economic and stock market growth while leaving the job market uncertain. (Axios)
• With details of President Trump's TikTok deal emerging, TikTok's new U.S. entity is expected to lease its algorithm from Chinese owner ByteDance. (Axios)
• Apple may be testing a robot or a new smart speaker without a display. (The Information)
• Exclusive: New wallet app gives users keys to their own personal data so they can control and profit off its use in training AI. (Axios)
• Exclusive: Employees across industries and job functions are racing to gain AI skills, according to new data from Udemy. (Axios)

There was a version of the Michigan Algorithm Decoder (MAD), an early programming language and compiler, that would print out a picture of Mad Magazine protagonist Alfred E. Neuman when a programming error was detected.