Today is Thursday, and as usual, that means there’s a lot to wrap up before Friday rolls around. So, let’s keep today’s newsletter on a positive note.
We might be witnessing the emergence of SeeDance 2.0. There’s been some buzz, particularly around cancellations. Sora, for instance, has stirred quite a conversation by replacing well-known copyrighted figures with historical memes.
For some context, SeeDance is a fresh AI video model capable of creating outlandishly unrealistic Hollywood-esque footage. Interestingly, Hollywood isn’t too pleased and has apparently sent a mountain of legal documents to its creator, ByteDance.
It would be fantastic if we could set up some standards, like MCP, to allow A. a computer to create lookalikes of individuals, and B. a straightforward way for people and companies to license themselves or famous characters to creators. This could benefit everyone involved. I mean, it’s essential to have no fakes, right?
So, what’s been happening in the AI sphere today?
- OpenAI is gearing up to access funds from a significant $100 billion investment.
- A glitch in Microsoft Office led to leaked confidential emails to the Copilot AI.
- World Labs secured a whopping $1 billion, with an additional $200 million from Autodesk for its 3D models.
- Google has rolled out new music generation features with its Gemini app using Lyria 3.
On another note, don’t miss the latest episode of the AI Fix podcast, featuring Corey Nolles. This week, he chats with Mark Stockley from MalwareBytes, diving into AI agents and the evolving tech landscape. It’s definitely something to check out if you’re into entertaining tech discussions.
Switching gears, on the topic of security and cryptocurrencies, OpenAI seems to demonstrate some concerning capabilities. Their recent coding agent, GPT 5.3-Codex, has an impressive—and somewhat alarming—72% chance of breaching vulnerable cryptocurrency smart contracts to siphon funds.
This collaboration with Paradigm has spawned a new benchmark tool, EVMbench, which tests AI’s efficiency in identifying, fixing, and exploiting security holes within smart contracts.
For those new to cryptocurrencies, smart contracts are like automated safes that follow specified rules in code. A flaw in this code can easily lead to a loss of funds, and once it’s gone, there’s no reversing that loss. I wonder—are there any smart contracts being written that utilize AI to interpret fixed rules before execution to sidestep this kind of issue?
Here’s the scoop on the benchmark results: GPT-5.3-codex achieved a score of 72.2%, capable of draining weak contracts’ funds roughly three-quarters of the time. For perspective, GPT-5 reached just 31.9% during a similar task six months ago. AI appears to excel in offensive moves rather than defensive ones. The best model only detected about 46% of vulnerabilities.
Interestingly, providing hints about where to investigate significantly boosts the effectiveness of bug fixes from 39% to an impressive 94%. It seems that the real challenge lies in the search process rather than in the ability itself.
But there’s more! In a wild case study, a GPT-5.2 agent executed a flash loan attack without any human guidance, wiping out the test vault’s entire balance in a single transaction. OpenAI is pushing this tool as a defensive weapon, backing it up with $10 million in API credits for cybersecurity researchers and expanding its Aardvark and Trusted Access for cyber professionals.
Now, why does this matter? Because the very AI that can help you draft emails can also drain your crypto vault in mere minutes. The race is on—will defenders implement these tools faster than attackers?
Also, Claude Sonnet 4.6 appears to be somewhat overengineered, with unnecessary features and explanations. To tackle this, here’s a simple guideline: “Do exactly what is asked. Pick one way and dive in. Avoid comparing options before proceeding.” This approach seems relevant for both coding and writing.
Before you go, have you checked out our latest episodes and articles? There’s a lot of content out there worth exploring, from insights into AI advancements to practical tips. Happy reading!
