Twitxy Saved

Nancy Wang

Verified account This account is verified. Learn more

@NancyZWang

🛡️ Cyber products at @OpenAI
👩🏻‍🏫 Founder @AdvancingWomeninTech
San Francisco, CA Joined March 2017
472 Following 1.9K Followers 476 Posts

Latest posts 5

Nancy Wang Verified account @NancyZWang
This week, the team launched GPT-6 Cyber.

There’s a big gap between finding a flaw and proving it’s exploitable. GPT-6 Cyber makes significant improvements here: it built working exploits for 948 Linux kernel bugs.

I’m excited to get this in the hands of our defenders 🚀

openai.com/index/introduc…
openai.com GPT‑6 Cyber: Advancing the frontier of aligned cybersecurity agents Introducing GPT‑6 Cyber, OpenAI’s cybersecurity model for finding exploitable vulnerabilities and developing fixes, trained to stay within authorized scope.
1 replies 0 reposts 11 likes 358 views View on X
Nancy Wang Verified account @NancyZWang
I shared our build vs. buy approach at the recent @MayfieldFund x @CitiVentures panel, moderated by @MattCarbonara.

@KFHightower, VP and CDAO at @StanfordHealth, explained Stanford's approach: it has the talent to build internally, while healthcare procurement cycles can outlast technology cycles. Kevin Smith, CIO at @lyft, is building internally too. Both looked to us to help make those systems safe to use.

When procurement takes longer than building, your prospect may ship a solution before they can buy yours.
Nancy Wang Verified account @NancyZWang
Agents should book your vacation without seeing your passwords.

With @1Password for Agents, your agent requests access to an account. You approve the request. 1Password handles the login.

1password.com/personal-ai-ag…
1password.com 1Password for Agents: Secure AI Password Manager | 1Password 1Password for Agents lets personal AI agents login without seeing your passwords. Connect supported agents and stay in control of login access.
1 replies 1 reposts 12 likes 533 views View on X
Nancy Wang Verified account @NancyZWang
Agents can gain more authority every time they succeed. Teams give them another tool, another permission. Each expansion of authority requires a new authorization decision.

I spoke with @BransPat for @CIOonline about how to address autonomy creep: authorize access at runtime, use short-lived credentials, and verify outcomes independently. Keep those controls outside the model.

Read Pat’s discussion with me:
cio.com/article/423570…
www.cio.com Autonomy creep: how AI quietly gains more authority than CIOs intend The biggest risk with agentic AI may not be deliberate autonomy but the gradual accumulation of permissions, tools, and decision-making power.