This briefing is informational only and does not constitute legal or financial advice. It reports on public industry developments and does not reference any Tienta client.
This Week in AI
Platform Roundup
- Anthropic (Claude): Anthropic now lets security teams run automated vulnerability scans powered by its most capable model, Claude Mythos 5, through a new "Claude Security" product — without giving them direct access to the model itself. If your IT vendor uses Claude, this may help it catch weaknesses faster.
- OpenAI (ChatGPT): OpenAI reaffirmed that it never stores or reviews prompts from eligible enterprise and API customers — a real gap from the free ChatGPT tier, which carries no such guarantee. Separately, ChatGPT logged about an hour of login and messaging problems on August 19, since resolved.
- Google (Gemini): Google is rolling out "Ask Gemini in Chat," an assistant built into Google Chat that can search your Gmail, Drive, and Calendar, draft messages, and summarize conversations without switching apps, starting August 26.
- Microsoft (Copilot): Microsoft patched a Copilot flaw, nicknamed "CoSnitch," that let one malicious link quietly pull data from a user's connected email, calendar, and cloud storage with no further clicks needed. No evidence of exploitation was found, but it's a reason to think twice before clicking unfamiliar Copilot-related links.
- xAI (Grok): A batch of Grok Lite users got nonsensical, garbled responses for part of last week; xAI called it "a rare temporary generation glitch," fixed for most by starting a new chat.
Beyond the Platforms
OpenAI slows down its next model after it crosses a cybersecurity danger line
OpenAI's upcoming "Astra" model showed cyberattack capabilities serious enough to trip the company's own safety threshold, so OpenAI paused development while it tightens security testing and monitoring for misuse. The added safeguards now consume roughly a fifth of the computing power used to run the model.
Why it matters: expect this kind of safety-driven slowdown occasionally if you rely on the most advanced AI models — it's the cost of using frontier tools, not a sign something's broken.
Anthropic moves toward a blockbuster public stock offering
Anthropic, the company behind Claude, has confidentially filed IPO paperwork and is expected to publicly file its prospectus by the end of August, following a private valuation of roughly $965 billion — a debut some analysts expect could rival SpaceX's.
Why it matters: a public listing would bring far more disclosure about Anthropic's finances — useful context if your firm relies on Claude and wants a clearer read on the company behind it.
Industry Spotlight
For Legal Practices
NYC Bar says lawyers should think twice before recording conversations, even with AI transcription tools
A new formal opinion from the New York City Bar Association says lawyers should, absent a good reason, avoid recording conversations with non-clients — including via AI note-taking tools — given the ethical and tactical risks, and should get consent from all parties when they do.
Why it matters: if your practice uses AI meeting recorders on calls with opposing counsel or witnesses, review when recording is actually necessary versus just convenient.
A Minnesota lawyer lost his law license for a month over AI-fabricated case citations
Attorney Faisal Ahmed agreed to a 30-day suspension after filing a 2025 brief citing court cases that, it turned out, didn't exist — the product of unchecked AI-generated content.
Why it matters: discipline for unverified AI citations is no longer rare; treat every AI-drafted citation as unconfirmed until you've pulled up the actual case.
Legal-tech vendors rolled out a wave of new AI tools ahead of the industry's biggest conference
Ahead of the ILTACON legal technology conference, vendors including DISCO, NetDocuments, Litera, and Clio announced new agentic AI features spanning e-discovery, document review, contract redlining, and client intake.
Why it matters: worth taking stock of what's newly available before committing to, or renewing, any single legal-tech vendor.
For Financial & CFO Firms
A major accounting firm is turning its own AI strategy into a service it sells to clients
Aprio, a top-25 accounting firm, is expanding its venture-investing arm to back more AI startups and rolling out "AI-as-a-service" consulting to help its clients build AI strategies and internal policies.
Why it matters: expect more professional-services vendors to offer AI advisory as a paid add-on — worth checking what's already in your existing engagements versus what would cost extra.
Financial firms' AI use is outpacing their governance, and regulators are watching
A survey of over 200 compliance officers found 84% of financial firms have AI tools in everyday use, but the typical firm applies AI in fewer than two of twenty surveyed functions — and regulators expect existing governance rules to apply to AI now, not once AI-specific rules arrive.
Why it matters: examiners are already asking about AI oversight and recordkeeping; a documented AI policy is cheap insurance against being caught flat-footed.
On the Horizon
A federal bill that would freeze state AI laws for three years is still just a proposal — but it's one to watch. Lawmakers released a 269-page bipartisan discussion draft this summer that would preempt state AI rules, including those in California, Colorado, and New York, for three years; it hasn't been formally introduced and has already drawn opposition from some of Congress's own AI policy leaders.
Why get ahead of this now: if it advances, it could simplify multi-state compliance — but that outcome is far from certain. Keep tracking state-level rules rather than pausing compliance work in anticipation.
Tax firms' AI use is compounding quickly, according to new survey data. A survey of over 1,000 tax professionals found weekly use of AI research tools nearly doubled to 60% over the past year, with another third of firms considering adoption soon, per Blue J and CPA.com.
Why get ahead of this now: as AI absorbs more routine research work, firms already shifting toward value-based or fixed fees will be better positioned than those still billing by the hour for work that takes a fraction of the time it used to.
Try This This Week
Before you or anyone on your team uses a free-tier AI tool — like free ChatGPT — for anything client-related, take five minutes to check that tool's data retention and training policy. This week's news that OpenAI's enterprise customers get a no-retention guarantee is a reminder that free and paid tiers often carry very different privacy terms. If you can't find a clear answer in the tool's settings or privacy page, treat that as the answer, and keep sensitive client material out of it until you can confirm otherwise.