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): CEO Dario Amodei published an essay, "We Must Pace the Frontier," calling on the industry to deliberately slow the rate of AI capability gains rather than pause development outright. Anthropic is unilaterally committing to give third-party evaluators permanent, employee-level access to its systems. More below.
- OpenAI (ChatGPT): Sam Altman agreed with Amodei publicly within hours ("I agree with Dario, we need to pace the frontier") and committed OpenAI to the same evaluator-access step, a notable moment between two companies that compete directly for the same enterprise customers.
- Google (Gemini): Google opened access to Claude Opus 5 to every engineer at the company through its internal Antigravity development platform, previously limited to select teams. Gemini remains, in Google's own words, the "primary and foundational model" for internal development.
- Microsoft (Copilot): Satya Nadella said Microsoft "welcomes" the "deliberate pacing needed to get alignment right" and published a Code of Conduct for Microsoft's own MAI models, opened for public consultation.
Beyond the Platforms
Anthropic's CEO asked the industry to slow down. Three rivals agreed, with one notable exception.
Amodei's essay argues AI labs should deliberately slow capability gains, by roughly one to two years industry-wide, so safety and alignment work has time to catch up, pointing to accelerating recursive self-improvement (models helping build the next generation of models) and August's OpenAI agent-swarm incident as the developments that changed his mind. Altman and Elon Musk ("Dario is right") backed the call within hours. Demis Hassabis, formerly of Google DeepMind, agreed with the goal but called the specifics "correct in direction" while arguing the details need work: he favors an independent standards body over labs individually granting evaluators access one deal at a time.
Why it matters: nothing here changes what you should do this week. What it signals is that the labs building these tools are now publicly disagreeing about how fast is too fast, in public, with specifics. That's useful context for any adoption-pace decision your firm is weighing, and it's the same argument this briefing keeps making about governance: the rules and safeguards are worth having before the pressure to move fast wins the argument internally too.
Industry Spotlight
For Legal Practices
A UK tribunal struck off a lawyer over AI-fabricated citations, and dishonesty was never even alleged
The Solicitors Disciplinary Tribunal for England and Wales banned Abhishek Kumar from the register of foreign lawyers after he twice filed submissions containing fabricated case citations generated by AI, first in a formal response and again in a follow-up email, in his own defense against the regulator prosecuting him. The tribunal never alleged he acted dishonestly. He was struck off anyway, based on how he first used the tool and what he did after the errors were pointed out.
Why it matters: this is a UK case, not US precedent, but the reasoning travels. A sanction doesn't require an intent to deceive. Careless reliance on an unverified AI citation, and a weak response once it's caught, is enough on its own.
For Financial & CFO Firms
The AICPA is pushing back on IRS guidance that CPAs must pass AI savings straight to clients
Office of Professional Responsibility guidance posted in June told tax practitioners using generative AI to reflect the resulting efficiencies in lower billing for reduced research and drafting time. The AICPA is now formally asking the IRS to clarify, arguing the language doesn't leave room for value-based pricing and ignores what firms actually spend on training, tooling, and governance to use AI responsibly. AICPA CEO Mark Koziel called the OPR guidance non-authoritative and "overstepping a little bit."
Why it matters: if your firm hasn't decided, on purpose, whether AI-driven efficiency shows up as margin or as an automatic client discount, this is the week to decide, before a regulator or a client assumes the answer for you.
On the Horizon
The industry now has two competing ideas for how AI safety commitments should actually be enforced. Amodei's plan has individual labs granting outside evaluators direct, permanent access to their own systems. Hassabis is pushing instead for a single independent standards body that all labs answer to. Which model wins matters for anyone relying on a lab's safety claims: a patchwork of one-off evaluator deals is a very different thing to verify than one shared standard.
Why get ahead of this now: this is the same question your own firm faces at a smaller scale, self-policing versus an outside standard, and it's worth having a real answer before a client or a regulator asks which one you're using.
Anthropic's public listing is still tracking toward a public S-1 filing in late September, with the investor roadshow unlikely to start before mid-October. No calendar date is set, and Anthropic has said the timeline depends on market conditions.
Why get ahead of this now: the public prospectus, whenever it lands, will be the first real audited look at a company a lot of firms now depend on for a core tool. Worth reading when it lands rather than relying on secondhand coverage of it.
Try This This Week
Look at the last AI-assisted task that saved your team real time this month, and ask one honest question: did that time savings become margin, or did it quietly become a discount nobody decided on? If a client noticed the work went faster and assumed the price should drop, that's a default, not a decision. Firms that get ahead of AI pricing now, the same way the AICPA is pushing the IRS to let them, will be negotiating from a policy instead of apologizing for one later.