In partnership with
News of the day
1. UK AI Security Institute finds GPT-5.5 comparable to Claude Mythos in cybersecurity vulnerability detection. GPT-5.5 is widely available. → Read more
2. Anthropic eyes a $900B+ valuation in a rapid funding round, potentially surpassing OpenAI and preparing for an IPO. → Read more
3. Apple's Mac Mini is sold out due to high demand for agentic AI tasks, with CEO Tim Cook expecting supply constraints for months. → Read more
4. Databricks leverages AI to help banks automate BCBS 239 compliance, reduce costs, and gain a strategic advantage over legacy systems. Future-proof your risk management. → Read more
Our take
Hi Dotikers!
May 1st, Labour Day. While France puts away its sprigs of lily of the valley and the terraces fill up, Dotika stays glued to the keyboard to feed our favourite subscribers' curiosity. We took a look at what LLMs do when no one's watching, and the timing is delicious.
Yesterday we watched Meta shut the open-source door to ship Muse Spark as a proprietary product. Today, the UK AI Security Institute releases its evaluation of GPT-5.5 on offensive cyber capabilities, and the picture gives us one more reason to keep these models under lock and key.
The verdict is unambiguous. GPT-5.5 hits 71.4% on AISI's Expert suite, ahead of Claude Mythos Preview at 68.6%, GPT-5.4 at 52.4%, and Opus 4.7 at 48.6%. More importantly, it's the second model after Mythos to solve The Last Ones end-to-end, a 32-step simulated corporate attack that takes a human expert about twenty hours to crack.
The chilling moment is buried in the appendix. On a Rust reverse-engineering challenge that had cost a Crystal Peak Security expert twelve hours, armed with Binary Ninja, GDB, and Z3, GPT-5.5 returned the correct answer in 10 minutes 22 seconds, for $1.73 in API costs. The model diagnosed a PIE relocations trap with no help, wrote its own disassembler, validated its emulator by cross-checking registers, fixed an interrupt inversion along the way, and solved the cryptographic constraint. This isn't so much a capability test as a snapshot of skill transfer.
AISI adds a footnote for the audience. Six hours of red teaming were enough for experts to develop a universal jailbreak on GPT-5.5's safety layer, and a configuration issue on OpenAI's side meant they couldn't verify the patch held.
Meanwhile, 43% of UK companies suffered a cyberattack in the last twelve months. The race between attackers and defenders just gained a new participant who works for $10 an hour, never sleeps, and obeys whoever pays the API.
Happy Labour Day to all of cybersecurity's human resources. You're going to need courage.
Alex.
Your business has grown. Is your accounting on the same path?
When you started out, doing your own books made sense. But the business you're running today isn't the one you started. If your accounting hasn't kept pace, it's quietly costing you — outdated financials, no clear view of what's actually profitable, and hours every week pulled away from the work that grows your business. At BELAY, our Financial Experts integrate directly into your business. They manage your books, reconcile accounts, run payroll, and deliver the timely insight you need to make big decisions with confidence. Stop guessing. Start knowing.
Meme of the day
