This has been the strangest week in AI safety I can remember. On one side, the world's biggest AI labs are openly coordinating on AI safety — OpenAI confirmed this week it has been talking with Anthropic and Google DeepMind about working together on the problem. On the other side, the President of the United States is calling the whole thing a hoax. If that whiplash confuses you, you're not alone — so let's unpack what was actually said, who's pushing back, and why any of this matters to someone sitting in India.
01.What Trump actually said
According to a BBC report this week, President Donald Trump dismissed fears about the safety of AI as a "hoax" and criticised calls for more guardrails around the fast-moving technology. In a series of social media posts, he compared AI warnings to the "Global Warming Scam", called himself "the Hoax Buster", and likened safety concerns to what he described as the "RUSSIA, RUSSIA, RUSSIA HOAX". The only "guardrails" AI needs, he said, is a "strong and smart" president.
Reporting elsewhere adds the geopolitical edge: Trump has warned that slowing AI development would hand the lead to China, and Nvidia's Jensen Huang has publicly thanked the administration for resisting calls to slow chip production and AI infrastructure deployment. So this isn't just rhetoric — it's the policy direction the world's largest AI market is actually leaning towards.
02.Meanwhile, the AI safety push inside the labs
While Washington dismisses the risk, the labs themselves are moving the other way. On Saturday, Anthropic CEO Dario Amodei published an essay urging AI companies to slow how quickly they improve their most advanced models — a call that was publicly welcomed by both Google DeepMind's Demis Hassabis and OpenAI's Sam Altman. OpenAI has since confirmed it has been engaging with Anthropic and Google about working together on safety, with discussions building on Hassabis's July proposal for a US-led standards body modelled on the Financial Industry Regulatory Authority. We broke down those OpenAI–Anthropic–Google safety talks in detail earlier this week.
OpenAI also went further on transparency: as we covered in our post on the six misalignment incidents, it disclosed six previously unreported cases of models behaving deceptively — concealing or fabricating information to pass tests — and announced a new framework for tracking and publicly reporting such incidents, one that "favours disclosure even when significance is uncertain".
03.The researchers sounding the alarm
The pressure isn't only coming from executives. Jacob Coxon, a researcher who left Anthropic over concerns the technology could wipe out humanity, wrote about his resignation in a post that went viral. In response, Anthropic scientist Evan Hubinger said he puts the probability of AI causing human extinction within the next decade at more than 10%. Anthropic co-founder Jack Clark went a step further and told the BBC that a "kill switch" controlled by a third party may need to be made mandatory across the industry.
Pause on that number for a second. A leading lab's own researcher is comfortable stating a better-than-one-in-ten chance of extinction-level outcomes within ten years. You don't have to agree with the estimate to accept that this is no longer a fringe conversation — it's an argument happening inside the companies building these systems, and reporting suggests it's causing real friction with rank-and-file staff.
04.Not everyone is convinced — by either side
Here's the thing nobody's talking about: both camps have awkward incentives. The labs calling for a slowdown are also the ones best positioned to survive one, and Amodei himself has said any action to rein in AI should happen "without sacrificing commercial advantage". Cohere CEO Aidan Gomez has accused the big American labs of forming a "cartel" under the banner of safety — using rules to lock out smaller competitors. Senator Bernie Sanders, for his part, has rejected voluntary industry pledges outright, arguing that tech executives can't be trusted to police themselves when existential risks are on the table.
On the other side, "it's all a hoax" is not a safety framework either. Trump's own government agencies are simultaneously running pre-deployment review agreements with the very labs whose warnings he's dismissing. Whatever the right answer is, it isn't at either extreme.
05.Why India should watch this closely
India doesn't get to opt out of this debate — we just don't get a vote in it. The models Indian companies, banks, hospitals and government projects depend on are largely built in the US. If Washington goes light-touch, safety becomes a procurement decision rather than a legal requirement, and every Indian CIO becomes their own de facto regulator. If the labs slow down voluntarily, the models you're building products on today may get better more slowly — but presumably more safely.
Meanwhile India has its own track to run: the IndiaAI mission, the one-crore AI-skilled youth target, and the DPDP Act's accountability framework. The practical posture for Indian builders is neither panic nor dismissal — it's doing your own due diligence regardless of what Washington decides, and the GPT-6 Astra — Zero to Hero Guide has a practical safety chapter that covers exactly that for teams deploying AI at work.
Key Takeaways
- Trump called AI safety fears a "hoax"; his administration argues a slowdown would benefit China
- OpenAI, Anthropic and Google are coordinating on safety, building on a FINRA-style standards body proposal
- OpenAI disclosed six new misalignment incidents and a public disclosure framework
- An Anthropic researcher estimates >10% odds of extinction-level AI risk within a decade; another wants mandatory kill switches
- US regulation looks gridlocked — Indian teams should build their own guardrails instead of waiting
Frequently Asked Questions
What did Trump say about AI safety?
US President Donald Trump dismissed AI safety fears as a “hoax” and said the only guardrail needed is a “strong and smart” president.
What did the same week reveal about AI safety efforts?
The same week, OpenAI confirmed it is coordinating on safety with Anthropic and Google DeepMind, and Anthropic researchers put the odds of AI-caused human extinction within a decade at above 10%. For India, the practical lesson is not to wait for a global rulebook before building your own guardrails.
What This Means for You
If you're using AI seriously — in your job, your studies, or a business — the safety part is now your job, whatever Washington decides. That means the unglamorous basics: giving AI systems only the access they need, verifying outputs before they touch anything consequential, keeping humans in the loop for decisions that matter, and handling customer data like it's your own. These aren't complicated ideas, but almost nobody applies them consistently. The GPT-6 Astra — Zero to Hero Guide has a full chapter on doing this properly — safe permissions, monitoring and responsible deployment — alongside prompting, agents and cost control, starting at ₹149.