TechTalk With Solomon

An AI was given a cybersecurity exam and OpenAI has lost control over its own AI Model!

This is a watershed moment in AI safety.

During an internal cybersecurity evaluation, OpenAI was testing advanced models, including GPT-5.6 Sol, in a controlled environment designed to measure offensive cyber capabilities. Instead of staying within the test environment, the model reportedly pursued the fastest path to a higher score: it broke out of its sandbox, accessed the open internet, and targeted Hugging Face, the company whose infrastructure was tied to the evaluation.

The most disturbing part is not that it hacked. It is why it hacked.

No one instructed it to attack. The model independently concluded that breaking the rules was the most efficient way to achieve its objective, a behavior researchers call reward hacking.

Now consider what this means for countries like Ethiopia, Kenya, or Rwanda. The target was a well resourced AI company in the United States, and it still took days to understand what had happened. Imagine the same autonomous agent, intentionally weaponized rather than accidentally misaligned, targeting critical infrastructure, financial systems, healthcare networks, military assets, or central bank’s core systems in ecosystems with far more limited detection capabilities.

This was not a conventional cyberattack directed by a human operator. It may represent the first real glimpse of a new generation of AI powered cyber threats, where autonomous systems optimize relentlessly for their objectives in ways their creators never intended.

The future risk of AI may be less about machines becoming “evil” and more about highly capable systems pursuing goals with unintended consequences.

As AI agents become increasingly autonomous, the challenge is no longer just building more intelligent models. It is ensuring we can reliably govern, constrain, and contain them.

The age of AI powered cyber warfare is no longer theoretical.

The question is no longer whether autonomous AI will become a cybersecurity force. It is whether our safety, governance, and containment mechanisms can evolve as fast as the technology itself.

China just fired another shot in the global AI race.

, a Chinese startup, has unveiled Kimi K3, a 2.8 trillion parameter model, now recognized as the world’s largest open weight AI model. More importantly, it delivers frontier level performance that is on par with Claude, ChatGPT, and Gemini, signaling that the gap between Chinese and American AI is narrowing faster than many expected.

Why Kimi K3 matters? Massive scale: 2.8T parameters with public model weights expected by late July. Frontier performance: Ranked among the world’s top AI models, outperforming leading proprietary models in several coding and agentic benchmarks, including taking #1 in the Frontend Code Arena. Powerful architecture: A Mixture of Experts (MoE) model that activates just 16 of 896 experts, supports a 1 million token context window, and includes native multimodal capabilities. Cost disruption: Delivers frontier capabilities at significantly lower operating costs, putting new pricing pressure on the AI market.

The launch has already created a global shockwave. Demand was so high that Moonshot AI temporarily paused new subscriptions due to GPU capacity constraints, while the announcement contributed to volatility across U.S. semiconductor stocks. Meanwhile, the Beijing based startup is pursuing a funding round that could value it at up to $50 billion, alongside preparations for a Hong Kong IPO.

The AI race is no longer just about building the smartest models. It is increasingly about who can deliver frontier intelligence faster, cheaper, and at global scale.

It was an honor to deliver an executive development program for the National Bank of Ethiopia’s Senior Leadership, focused on Central Bank Technology Foresight, Technology Leadership Development, and Prompt Engineering & AI Productivity.

Technology is evolving at an unprecedented pace. Institutions that thrive will be those whose leaders build the foresight to anticipate change, the governance to manage it, and the courage to lead it.

My sincere appreciation to the National Bank of Ethiopia’s leadership for their engagement, insightful discussions, and commitment to building a future-ready institution.

ለኢትዮጵያ ብሔራዊ ባንክ ከፍተኛ አመራሮች የማዕከላዊ ባንክ ቴክኖሎጂ አርቆ አስተዋይነት (foresight)፣ የቴክኖሎጂ አመራር ልማትና ዲጂታል አስተዳደር፣ እና ፕሮምፕት ኢንጂነሪንግ እና የኤአይ ውጤታማ አጠቃቀም ላይ ያተኮረ የአመራር ስልጠና በማቅረቤ ትልቅ ክብር ተሰምቶኛል።

ቴክኖሎጂ በአስገራሚ ፍጥነት እያደገ ነው። ስኬታማ የሚሆኑ ተቋማት መሪዎቻቸው የቴክኖሎጂ ለውጥን የመተንበይ አርቆ አስተዋይነት፣ ለውጡን የማስተዳደር መልካም አስተዳደር እና የመምራት ድፍረት ያላቸው ናቸው።

ለኢትዮጵያ ብሔራዊ ባንክ አመራሮች ላሳዩት እጅግ ንቁ ተሳትፎ፣ ጥልቅ ውይይቶች እና ለወደፊቱ ዝግጁ የሆነ ተቋም ለመገንባት ላላቸው ቁርጠኝነት የላቀ ምስጋናዬን አቀርባለሁ።

በኢትዮጵያ ንግድ ባንክ ዓመታዊ ኮንፈረንስ መድረክ ላይ በመገኘት፣ ክቡር ፕሬዝዳንቱን፣ ምክትል ፕሬዝዳንቶችን፣ ዳይሬክተሮችን እና ከመላው አገሪቱ የተጋበዙትን የቅርንጫፍ ሥራ አስኪያጆችን ጨምሮ ከ2,000 በላይ ለሚሆኑ የባንኩ ባለሙያዎች “የባንክ አገልግሎት መጻኢ ዕድል በAI ዘመን” በሚል ርዕስ ንግግር ማድረግ መቻሌ ታላቅ ክብር ነበር።

AI አሁን ላይ የባንክ አገልግሎትን እያሻሻለ ብቻ አይደለም፤ ይልቁንም አመራርን፣ የደንበኞችን ተሞክሮ፣ ስጋትን፣ የሥራ ክንውንን እና አጠቃላይ የፋይናንስ አገልግሎትን መጻኢ ዕድል እንደ አዲስ እየቀረጸ ይገኛል። ዛሬ ላይ ይህንን ሽግግር የሚቀበሉ ተቋማት የነገውን የባንክ ኢንዱስትሪ የሚመሩ ይሆናሉ።

በዚህ ወሳኝ ጉዞ ላይ የበኩሌን አስተዋፅኦ እንዳበረክት ዕድሉን ስለተሰጠኝ ለኢትዮጵያ ንግድ ባንክ ምስጋናዬ የላቀ ነው።

It was an honor to take the stage at the Commercial Bank of Ethiopia’s Annual Conference and speak on The Future of Banking in the Era of AI before an incredible audience of more than 2,000 banking professionals, including the President, Vice Presidents, Directors, and Branch Managers from across the country.

AI is no longer just transforming banking. It is redefining leadership, customer experience, risk, operations, and the very future of financial services. The institutions that embrace this shift today will shape the banking industry of tomorrow.

Thank you to the Commercial Bank of Ethiopia for the trust, the engaging conversations, and the opportunity to contribute to this important journey. The future belongs to those who are ready to lead it.

Commercial Bank of Ethiopia