Anthropic, the brains behind Claude AI, recently announced a breakthrough. Their model has designed working mini-proteins that could form the basis of new drugs—something that usually takes specialists months to do.

Claude nailed it for 14 out of 15 cases, according to Anthropic. These proteins can bind to disease-causing pathogens and neutralize them—a key step in drug development.

But the numbers here come straight from Anthropic themselves—no independent experts have verified this yet. That said, Claude's success rate ranges from 22% to 35%, which is impressive compared to typical industry standards of 10-15%. Some designs even outperformed previously published results.

Interestingly, the physical lab work wasn't handled by Anthropic either. Two external parties synthesized and tested these proteins. The whole process still takes weeks—no matter how smart the software is.

The Dull Work Gets Faster Too

In a separate trial, Claude's Opus 5 model could read chemical measurements in just 25 minutes. That’s a job usually done by chemists manually over hours—and even took four days for an external lab this time around.

Opus 5 even corrected its own mistake during the process—an impressive feat that you’d typically expect from a meticulous colleague rather than an AI system.

A Personal Quest

This work is more than just about numbers. Anthropic’s CEO, Dario Amodei, who lost his father to hepatitis C before effective treatment was available, sees this as a chance for AI to help cure many diseases within five to ten years.

But he also recently toned down expectations on X, noting that the industry hasn’t yet delivered on its big promises. Real progress in treating diseases, not slick marketing campaigns, will earn back trust.

The Dark Side of Progress

There's a flip side to this innovation. Anthropic highlights how the same technology that could speed up drug development could also be used for nefarious purposes like bioweapons creation. To avoid misuse, they restrict access to their most powerful models only to screened researchers.

The Long Road Ahead

While these results are exciting, it's worth noting that designing a protein is just the first step towards creating an actual drug. Anthropic acknowledges this too. The journey from lab bench to pharmacy counter is still long and winding—but AI is definitely making strides.

Robbert
Robbert

Continue Reading

News 3 min read

Zetrix AI's governance crisis deepens as its entire audit committee walks out

Zetrix AI's troubles are escalating fast. Just days after a credit rating agency flagged the company over a covenant breach on its sukuk, the Malaysian blockchain and AI firm has now lost its entire audit committee, with most of its board resigning in quick succession. The departures leave the listed company without key oversight functions at exactly the moment its creditors and regulators are watching closest, and come on the heels of a second rating downgrade in two weeks that pushed the sukuk closer to default territory. Investors and bondholders are left asking who is steering the company and whether more resignations or disclosures are coming.

Unknown
News 3 min read

Customs intercepts RM50 million in AI servers over suspected export breach

Malaysian customs officials say they have seized a shipment of roughly 30 servers packed with advanced AI chips, worth close to RM50 million, over a suspected breach of export-control rules. The seizure is one of the biggest of its kind flagged publicly by Malaysian authorities this year, and comes as the country tries to balance its ambitions as a regional AI and data centre hub with growing scrutiny from Washington over where high-end chips end up. Officials have not yet named the buyer, seller or intended destination of the hardware, and investigations are ongoing. The case adds to a string of enforcement actions around advanced chip flows through Malaysia in recent months, as regulators try to show they are serious about plugging loopholes without scaring off legitimate investment.

Unknown
News 3 min read

Malaysia's dedicated AI island is taking shape, here's what's being built

Plans for a dedicated Malaysian zone built around AI and data centre infrastructure are progressing, with land, power and connectivity groundwork underway. The report details which agencies and developers are involved, the scale of planned facilities, and the incentives being offered to attract AI and cloud tenants. Unlike earlier announcements, this piece focuses on construction and infrastructure milestones rather than investment pledges, giving a clearer picture of how close the project is to operational. Readers tracking Malaysia's broader AI infrastructure ambitions should note how this zone fits alongside other data centre clusters in Johor and the Klang Valley, and whether power supply commitments are keeping pace with demand.

Unknown